Before you can analyse anything, you need data. The good news is that football has more free public data than almost any other sport. The bad news is that it comes from a dozen different sources with inconsistent coverage, opaque licensing, and access rules that differ from host to host. Here is every source worth your time, ranked on what we measured rather than on opinion: what each one let an automated pull do on the day we checked, what we actually got from it, and the quirks we hit.

Free soccer data sources ranked on what we measured

Most rankings of free football data are opinions about usefulness. This one is built from three things we could check: what each source’s robots.txt told an automated agent when we read it on 29 October 2026, what we hold from each source, counted from the files on disk, and the quirks our pull scripts recorded. The criteria, applied in this order:

  1. Access on the check date. The robots.txt answer to this site’s named pull agent (one request per host on 29 October 2026, nothing else requested), plus the request behaviour our pull scripts recorded.
  2. Data we hold and could count. Only a source we have pulled can be measured, so a pulled source ranks above one we have not pulled.
  3. Granularity of what we hold. Event-level data beats per-team match statistics, which beat results only.
  4. Quirks we measured. Duplicates, gaps and query changes, counted.

A source we never pulled cannot be scored on the last three, so it sits below every source we did pull, ordered by access alone. Where access is the same, the sources share a rank.

Seven free soccer data sources, ranked on measured facts. robots.txt answers were read on 29 October 2026 with this site’s named agent; other dates are given where they differ. StatsBomb counts are from our census of 20 October 2026 (counts only); ESPN counts are from the files on disk, with their pull dates.
Rank Source What we could get Coverage we counted Quirks we hit robots.txt / access
1StatsBomb open dataEvent files: every pass, shot and carry, with xG on every shot3,961 matches in 80 competition-seasons (24 competitions), 24 June 1958 to 27 July 2025; 426 matches with 360 freeze frames; 1,310 women’s matchesCurated: 29 competition-seasons are a single match and 26 (670 matches) follow one team; 418 Premier League matches, 380 of them from 2015/16raw.githubusercontent.com 404: no robots.txt, so no rules for the files; github.com 200, repository pages open
2ESPN public APIResults, goal and red-card minutes, per-team match statistics9,170 Premier League matches: all 24 seasons from 2002-03 to 2025-26, plus 50 of 2026-27 (pulled 5 Oct 2026); 2025-26 box scores for 380 of 380 matches; 89 completed 2026-27 matches (1 Nov 2026); World Cup 2026, 104 of 104 (19 Jul 2026)9,527 rows as pulled: 2009-10 listed twice, 357 extra rows; 12 goal lists that do not add up to the score; 51 of 292 monthly requests empty; 9 statistics per team after 2025-26, against 28; date-range query answers 400 (since 18 Sep 2026)site.api.espn.com 403 to our named agent, even for robots.txt; custom User-Agent strings answered 403 (16 Aug 2026), so our pulls send Python’s standard library agent
3=OpenFootball / jokecampNot pulledNone countedNone measuredgithub.com 200: open to generic agents; a group for 5 named AI crawlers closes 162 paths (including /*/raw/) with a one-second crawl delay; raw.githubusercontent.com 404
3=Kaggle datasetsNot pulledNone countedNone measuredwww.kaggle.com 404: no robots.txt at this host
5football-data.co.ukNot pulledNone countedNone measured200: open to generic agents (Disallow: left empty), but Disallow: / for 12 named AI and scraping agents
6=UnderstatNot pulled (this site retired its Understat pull)None countedNone measured200: User-agent: * / Disallow: /, the whole site closed to every automated agent
6=FBrefNot pulledNone countedNone measured403 to our named agent for robots.txt itself; nothing further requested

Why StatsBomb is first and ESPN second. StatsBomb’s free release is the only source here that gave us event-level data with no access condition attached: the raw files sit on a host with no robots.txt, and the repository pages are open. Its limit is curation, and the census shows how much. Of its 80 competition-seasons, 29 are a single match (22 of them finals) and 26, holding 670 matches, follow one team through a season. Only four seasons, all from 2015/16, have 377 or more matches. ESPN goes much deeper on English football (9,170 Premier League matches against StatsBomb’s 418), but it is per-team statistics rather than events, and it carries the most quirks we measured. The 2009-10 season is listed twice, adding 357 rows, and 337 of those carry a different score from the real fixture. Twelve matches have a goal list that does not match the final score. And the feed changed under us twice: custom User-Agent strings got 403 when we tested on 16 August 2026, and the date-range form of the scoreboard query has answered 400 since 18 September 2026, so every pull walks the calendar one month at a time. Our saved ESPN files keep only scores, event minutes, team statistics and fixture details.

The middle: open, but nothing we could count. Nothing in the robots.txt of GitHub (where the OpenFootball and jokecamp files live) or Kaggle stopped a generic agent on 29 October. We hold no files from either, so we cannot vouch for their coverage, and many Kaggle uploads are re-exports of other sources on this list, which brings those sources’ terms along with them. football-data.co.uk ranks below them for one measured reason: its robots.txt is open to generic crawlers but names 12 AI and scraping agents and closes the whole site to them. We have not pulled from it.

The bottom: closed on the check date. Understat’s robots.txt is two lines that close the whole site to every automated agent. FBref answered our agent with 403 before any rule could be read, and we requested nothing else from it. Both publish data many people want, xG and shot records at Understat and Opta-powered advanced statistics at FBref, but on 29 October 2026 neither was open to an automated pull that follows robots.txt. That is why this site no longer pulls from Understat, and why its recent Premier League and 2026 World Cup figures come from ESPN’s public API and its 2022 World Cup work from StatsBomb’s files.

Method and limits. Each robots.txt was requested once per host on 29 October 2026 by our robots.txt check, with the agent SoccerAnalytics-data/1.0, and the answers are saved with the date. A robots.txt answer is a snapshot. Hosts change their files, and an answer to our agent can differ from the answer to yours: both 403s were given to our named agent. The ESPN counts come from the files behind this site’s own figures, de-duplicated as described below; the StatsBomb counts come from the release’s own match lists, read on 20 October 2026, and copy no StatsBomb rows. Club and league official APIs were not checked, so they are not ranked. This ranking describes what an automated pull could do on those dates. It is not legal advice, and it is not a reading of anyone’s terms of use: robots.txt speaks to automated agents, while a site’s terms speak to everyone, so read both before you build on a source.

Reproduce it. Every count in this section and the ranking table are written to /data/stats/every-free-public-soccer-data-source-ranked.json by our analysis script. It reads only files already on disk (the saved robots.txt answers, the StatsBomb census and the ESPN files under /data_layer/) and makes no requests.

What each source offers

The table above ranks what we measured. This one describes what each source says it offers, in the same order, with the catch that comes with each. The descriptions of scope are the sources’ own, not our counts.

Free public soccer data sources, in measured rank order. "Access" describes how the data is reached; access status is as recorded on 29 October 2026.
Source What you get Access The catch
StatsBomb open data Full event streams: every pass, shot, carry, pressure, with freeze frames and pre-computed xG. Dozens of competitions including multiple full seasons of WSL, La Liga, Champions League and the last several World Cups. Git clone or statsbombpy Python client; no account needed. Competition coverage is curated by StatsBomb's commercial interests; the Premier League is almost entirely absent. Non-commercial use only; attribution required. Data can be large: the full repo is several GB.
OpenFootball / jokecamp Results, fixtures, and standings for major leagues and international tournaments, formatted as CSV or JSON and hosted on GitHub. Git clone or raw GitHub download. Community-maintained with no formal update guarantee. Coverage is uneven: some repositories are years out of date. Good for historical lookups; not a live feed.
Kaggle datasets A long tail of community-uploaded datasets: historical results, player attributes from Football Manager–style databases, pre-built xG tables, and occasional scrapes of the sources above. Kaggle API or browser download. Quality ranges from excellent to completely wrong. Always check the source of the source: many Kaggle datasets are re-exports from FBref or Understat, so the same licence restrictions apply. Treat as a discovery layer, not a primary source.
football-data.co.uk Match results, final scores, half-time scores and referee for the top European and some lower-division leagues, going back to the early 1990s in places. Plain CSV files. CSV download from the site. Its robots.txt treats generic agents and 12 named agents differently, so check which group your agent falls in before automating anything. No event data, no player-level data: purely match-level results. Its robots.txt is open to generic crawlers but shuts out 12 named AI and scraping agents (read 29 October 2026). We hold none of its files.
Understat Match xG, shot-level xG with location, and season xG tables for the top six European leagues: Premier League, La Liga, Bundesliga, Serie A, Ligue 1, and RFPL. Data goes back to the 2014/15 season. The website only; there is no API. Its robots.txt closes the whole site to automated agents, so no automated route follows its rules. No formal API and no explicit open licence. The underlying xG model is a black box: the exact feature set is undocumented. RFPL coverage can lag during the season. Its robots.txt closes the whole site to every automated agent (read 29 October 2026).
FBref Broad advanced stats for hundreds of competitions: xG, progressive passes and carries, defensive pressure, squad and player-level. Powered by Opta data. The website only; no official API. The terms of service restrict systematic scraping, and the data is Opta-licensed, so redistribution is off the table. On 29 October 2026 it refused our pull agent with HTTP 403, even for robots.txt.
Club and league official APIs (not ranked: not checked) A handful of leagues publish official APIs: the Premier League has a documented public API; some national federations have fixture feeds. Coverage is results and lineups only. HTTP REST API, no client libraries; usually no authentication required for basic endpoints. These APIs are undocumented, unofficial, and change without warning. The Premier League's API in particular has broken third-party tools several times. Treat as a pleasant surprise rather than a dependency.

What we measured on 29 October 2026

Two things can be checked rather than judged. The first is what each source tells an automated agent. On 29 October 2026 we requested robots.txt once from each host below, with this site’s named pull agent, and nothing else; the answers are saved with the date. The second is what the files behind this site’s own figures actually contain, counted from the copies on disk.

robots.txt of each source, read once per host on 29 October 2026 with this site’s named pull agent (one request per host, nothing else requested).
Source Host checked Answer What it says to an automated agent
StatsBomb open data (files)raw.githubusercontent.com404No robots.txt, so no rules for the raw JSON files.
StatsBomb, OpenFootball, jokecamp (repository pages)github.com200Open to generic agents. A group for five named AI crawlers asks for a one-second crawl delay and closes 162 paths, such as /*/raw/ and /*/blame/; repository front pages stay open.
Understatunderstat.com200User-agent: * / Disallow: /: the whole site is closed to every automated agent.
FBreffbref.com403Refused our agent before any rules could be read; nothing further was requested.
football-data.co.ukwww.football-data.co.uk200Open to generic agents (Disallow: left empty), but 12 named AI and scraping agents get Disallow: /.
Kaggle datasetswww.kaggle.com404No robots.txt at this host.
ESPN public API (this site’s results source)site.api.espn.com403Refused our named agent, as the API does any custom User-Agent; the pulls behind this site’s ESPN files send Python’s standard library agent instead.

These answers set the bottom of the ranking. Understat closes its whole site, and FBref turned our agent away before we could read its rules. football-data.co.uk is open to generic crawlers but shuts out a named list of AI and scraping agents. We pull from none of the three.

What our ESPN files hold

Coverage claims are easy to make and rarely counted, so here are the counts for the ESPN files this site serves under /data_layer/, with the gaps we found in them.

What the ESPN files behind this site contain, counted from the saved copies (pull dates as recorded in each file).
File (pulled) What it holds Gaps and quirks we measured
Premier League results with goal and red-card minutes (5 Oct 2026)9,170 matches: every 380-match season from 2002-03 to 2025-26, plus 50 matches of 2026-27. 24,981 goals with a minute (1,740 penalties, 894 own goals) and 1,340 red cards with a minute.9,527 rows as pulled: ESPN lists 2009-10 twice, and 337 of its 357 extra rows carry a different score from the real fixture. 12 matches, all before 2017-18, have a goal list that does not add up to the score. 51 of 292 monthly requests returned no completed match: 47 June or July close seasons, April, May and August 2020, and October 2026 before its first game.
Added time, 2016-17 onward (5 Oct 2026)3,850 matches with the minute each half ended.The added time the fourth official announced is there for 823 matches and missing for 3,027; one match summary has no end time for its second half.
2025-26 box scores (7 Sep 2026)380 of 380 matches, 28 statistics per team.No missing match summaries; every score matches the separate 2025-26 results file row for row.
2026-27 to date (1 Nov 2026)89 completed matches, of 380 fixtures listed.9 statistics per team, against 28 in the 2025-26 file.
World Cup 2026 (19 Jul 2026)104 of 104 matches, 9 statistics per team.Pass accuracy arrives as a fraction rounded to one decimal: across 208 team-matches it takes only five values, 0.5 to 0.9, too coarse to separate teams.

Two more quirks live in our pull scripts’ notes rather than in the files. ESPN stopped accepting the date-range form of its scoreboard query (it answered HTTP 400 when we checked on 18 September 2026), so every pull here walks the calendar one month at a time. And it refused custom User-Agent strings when we tested on 16 August 2026, which matches the 403 above. For StatsBomb’s free release, our census of 20 October 2026 counted 3,961 matches across 80 competition-seasons, the latest dated 27 July 2025, and 418 Premier League matches, 380 of them from 2015/16; the full breakdown is in getting started with StatsBomb open data.

1st: StatsBomb open data, the only free event data we could count

Nothing else comes close on granularity. StatsBomb's free dataset is the same format as their commercial product: every event on the pitch, timestamped and coordinated on a 120×80 grid, with optional freeze-frame snapshots of every player's position at the moment of each shot. The pre-computed xG values are baked in. The statsbombpy Python client means you are three lines of code from a full match worth of events, with no account, no key, and no rate limit to worry about.

The honest caveat is that StatsBomb's free release is shaped by their marketing. Competitions that their commercial clients care about (the Women's Super League, the African Cup of Nations, La Liga with its Messi years) are well represented. The Premier League, where every club already pays for data, barely appears: 418 matches in our census, 380 of them from the single 2015/16 season. If your analysis lives and dies in the English top flight, StatsBomb open data is a research tool, not a production one.

Licence summary
Non-commercial use with attribution. Republishing raw data files is not permitted. Building public visualisations and writing articles is fine; selling a product built on the data is not. When in doubt, read the repository's README.

2nd: ESPN’s public API, the deepest results file we hold

ESPN’s public scoreboard and match-summary feeds are where this site’s Premier League and 2026 World Cup figures come from: 24 full Premier League seasons with goal and red-card minutes, a full season of 28-statistic box scores, and every World Cup 2026 match. It ranks second because what it gives is per-team statistics rather than events, and because of the access conditions and quirks counted above. Treat any re-pull as something that can break without notice: the date-range query and custom User-Agent strings both stopped working during 2026.

Joint 3rd: OpenFootball and Kaggle, open but uncounted

The various OpenFootball repositories on GitHub, including the long-running jokecamp collection, are excellent for structured historical results when you want clean JSON or CSV without any scraping at all. The problem is maintenance: open-source sports data projects are notorious for going quiet at exactly the moment you need last season's data. Before building a pipeline on any of them, check the last commit date.

Kaggle sits above these in discoverability but below them in trustworthiness. It is where you find that someone has already assembled the dataset you need, and also where you find out, three hours later, that they made a systematic error in the season mapping. Use it as a signpost to what is possible and then go upstream to the original source.

5th: football-data.co.uk, a results archive with named-agent rules

Correction, 29 October 2026
This section first named Tony Hirst as the person behind football-data.co.uk. The site is Joseph Buchdahl's.

Joseph Buchdahl's site has been publishing plain CSV files of match results since the late 1990s, and it still works the same way: a season of a league is one CSV file. The historical depth is unmatched: some English lower-division series go back thirty years.

Its place in our ranking reflects what we could measure, not its quality as an archive. We hold none of its files, and its robots.txt, read on 29 October 2026, is open to generic crawlers but closes the site to 12 named AI and scraping agents. As a resource, it is a results archive: there is no pass-level data, no xG, no player-level anything. If your question is "do home teams in League One win more in November than March?", it is the obvious place to look. If your question involves anything that happened during the ninety minutes, you need a different source.

Joint 6th: Understat, closed to every automated agent

Understat built its xG model independently and has been running it since 2014. For the six leagues it covers, it shows not just season totals but the individual shot record: location, xG value, minute, and outcome. The long-standing weakness is the closed model: you cannot inspect what Understat's xG formula actually weights, which matters for any published methodology.

The bigger problem now is access. Its robots.txt, read on 29 October 2026, is two lines, User-agent: * and Disallow: /, which close the whole site to every automated agent. That rules out the scraping libraries that read the data embedded in its pages, the route this page used to describe, and it is why this site retired its own Understat pull. For shot-level xG on the free tier, StatsBomb's event files carry an xG value on every shot, within the competitions it releases.

Joint 6th: FBref, breadth we could not reach

FBref is remarkable in scope. Hundreds of competitions, player-level advanced stats, season-on-season trends going back years, powered by Opta, which means the underlying data quality is as good as anything in the commercial world. For competition breadth it has no peer among free sources.

The friction is access. On 29 October 2026 fbref.com answered our named agent with HTTP 403 for robots.txt itself, so we could not read its rules, and we requested nothing else. Its owner's terms of service also restrict systematic scraping, and the Opta licence means you cannot republish the underlying numbers. If you use FBref, use it the way its owners allow: read the terms and the robots.txt first, and if the host turns your agent away, take that as the answer.

Following a major tournament on free data

This section condenses a separate article first published on 10 June 2026.

A World Cup is a different job from a league: 104 matches across a single month, between international sides you may have little prior data on. It helps to separate the two kinds of data you will reach for, because no single source does both equally well. Live and recent match data (shots, possession and per-team stats as matches finish) is what you use to ask whether a 1–0 was deserved; speed and coverage matter. This site followed the 2026 tournament on ESPN’s public scoreboard and match-summary feeds, which hold all 104 matches with 9 statistics per team; FBref, often suggested for this job, refused our agent when we checked on 29 October 2026. Historical context, the event-level record of past World Cups, is what grounds a take (“is this xG total high for a World Cup match?”) in something other than vibes; depth matters more than freshness, and StatsBomb’s open release of past World Cups is the single best free resource for it, though it is curated and historical, not a live feed.

Free versus paid, honestly. Free gets you a long way: reading xG, shot data and team stats after matches, the complete event-level history of past World Cups, and results archives. What money buys is real-time event streams, the complete current-event dataset rather than a curated historical slice, and the licence to build commercial products on top. Professional analysts and clubs pay for those things because their job demands them; a fan, a writer or a hobbyist building charts for a blog generally does not need any of it.

Before the 2026 tournament, we set four expectations for following it on the free tier. Data lags by hours, not seconds, sometimes longer on a packed matchday when many games finish close together; the data is for the morning-after analysis. Coverage of a major tournament is strong, with none of the patchy-coverage problem of obscure club competitions. Models differ, so cross-check: different providers run different xG models, so a match’s xG total can vary depending on whose number you read; note which source a figure came from and do not treat two providers’ numbers as interchangeable to the decimal. And FIFA is the source of record for the facts (fixtures, results, squads and the official format), not an analytics source, but the place to confirm what is actually true before you build a number on top of it.

A note on licences and rate limits

Two rules that are worth tattooing on your wrist before you start any project that touches free public data.

First, "free to access" is not the same as "free to use". StatsBomb's open data requires attribution and prohibits commercial use. FBref's data is Opta's property, licensed to Sports Reference for display: scraping it does not transfer any licence to you. football-data.co.uk does not publish a formal licence at all, which in practice means you cannot assume you have any rights beyond personal use. Whenever you plan to publish derived work or share data files with others, read the terms. If there are no terms, ask before you ship.

Second, robots.txt and rate limits are not suggestions. Read a host’s robots.txt before an automated pull, and stop if it closes the paths you want or turns your agent away; a 403 is an answer, not a puzzle to solve with another User-Agent. Where a host does allow automated requests, a polite pull caches responses and sleeps between requests (ours keep at least a second apart, or the host’s crawl delay if it sets a longer one). An impolite one burns through a site’s infrastructure and gets blocked, and the community’s continued access to free data depends on pulls behaving themselves.

The data behind this piece

  • Measured on this page: the robots.txt answers of 29 October 2026 (saved per host, with a summary, by our robots.txt check) and the ESPN files served under /data_layer/, counted by our archive loader; StatsBomb counts from our 20 October 2026 census of its competition list and match files (counts only). The ranking and its counts: /data/stats/every-free-public-soccer-data-source-ranked.json, written by our analysis script.
  • StatsBomb open data: the richest free event-level dataset in football.
  • StatsBomb: documentation on the data spec, event types, and freeze-frame format.
  • Understat: xG and shot-level data for the top six European leagues since 2014/15; its robots.txt closed the whole site to automated agents on 29 October 2026.
  • FBref: broad advanced stats across hundreds of competitions, powered by Opta.
  • football-data.co.uk: match results going back decades, in plain CSV.
  • jokecamp FootballData: community-curated results and fixture data on GitHub.
  • Kaggle datasets: community-uploaded football datasets; useful as a discovery layer.
  • soccerdata: a Python library with readers for FBref, Understat and others; check each source’s robots.txt and terms before using any of them (Understat’s closed the whole site on 29 October 2026, and FBref refused our agent).

Published by SoccerAnalytics from public data and the sources named in this piece. How this site is made.