Open Data Directory
High-quality, non-corporate data sources from academic institutions, governments, and non-profits — curated for researchers, analysts, and builders.
No sources match — try a different filter or search term.
Know a source that belongs here?
Suggest a source →Quick vetting checklist
Before trusting any data source, run through these checks. They take about two minutes and will save you from building on shaky foundations.
1
Check the domain
.gov (government), .edu (university), and .org (non-profit) are generally safer than .com. Not all .orgs are trustworthy, but if it's a .com leading with "enterprise solutions" or requiring a business email to access anything, be skeptical.
2
Look for a DOI
Reputable datasets almost always carry a Digital Object Identifier — a permanent academic ID that looks like 10.5281/zenodo.123456. No DOI on a research dataset is a yellow flag worth investigating.
3
Check the license
Look for CC0 (public domain — fully free to use) or CC BY (free with attribution). These allow reuse without fees or legal risk. Avoid sources with no stated license or vague "terms of use" that could restrict your work downstream.
4
Read the About page
If it talks about "research," "science," and "open access" — good sign. If it leads with "solutions," "partners," or "enterprise" — proceed with caution. The most important question: who funds it, and do they have an interest in the findings?
5
Check for methodology documentation
A trustworthy dataset explains how the data was collected, by whom, when, and what its known limitations are. No methodology section means you cannot assess quality or bias — which means you cannot responsibly use it.
6
Watch for red flags
Be wary of: data that perfectly supports a company's own product claims, no named authors or institutions, no date or update history, "proprietary methodology" with no explanation, and paywalled previews designed to upsell you on the full dataset.
Vetting in the wild
You'll often encounter data sources through social posts, articles, or someone citing a statistic. Here's how to trace it back to its origin before using it.
→
Always find the primary source
Never trust a statistic from the article that cited it — go upstream. Search for the original study, report, or dataset. If the article doesn't link directly to it, that's already a problem worth noting.
→
Check re3data or DataONE for unknown repositories
If someone cites a repository you don't recognize, look it up on re3data.org — the global registry of vetted research repositories. If it isn't listed there, research it before trusting it.
→
Check who funded the research
Funding source matters. NSF, NIH, EU Horizon, and government grants require conflict-of-interest disclosure. Corporate-funded research often does not. Always check the paper's acknowledgments section before treating findings as neutral.
→
"Available on request" is a red flag
If a paper lists its data as "available upon reasonable request" rather than linking to a named repository, the data may be lost, restricted, or never properly organized for sharing. Prioritize papers with named, accessible repositories.
How much data do you actually need?
A common trap in data projects: downloading everything and figuring out what you need later. Estimating your requirements upfront saves time, storage, and scope creep.
| Task type | Typical data volume | What matters most |
|---|---|---|
| Exploratory analysis | A representative sample (5–10%) is usually enough to start | Coverage, not completeness |
| Training a model | Depends heavily on task — classification needs far less than generation | Quality over quantity; label accuracy |
| Statistical research | Determined by required statistical power — often less than you think | Methodology documentation, sampling method |
| Benchmarking | Standardized test sets (often fixed sizes) | Reproducibility, established baseline |
| Policy or journalism | Whatever the primary source published | Provenance and recency |
Rule of thumb: if you can clearly state your research question, you can usually define the minimum data you need to answer it. Start small, validate your approach, then scale.
What file formats signal about a dataset
The format a dataset comes in can tell you a lot about how it was intended to be used — and how much cleanup work lies ahead.
| Format | What it signals | Watch out for |
|---|---|---|
| CSV / TSV | Research-ready, widely interoperable — generally a good sign | No schema docs, encoding issues |
| JSON / JSON-LD | Modern and structured — common in APIs and linked data | Nested structures that require flattening |
| Parquet / Arrow | Optimized for large-scale analysis — serious, intentional data | Requires specific tooling (pandas, Spark, DuckDB) |
| Excel (.xlsx) | Often manually curated — may have formatting embedded as data | Merged cells, color-coded meaning, hidden rows |
| PDF tables | Data not intended for reuse — extracted from reports | Significant cleaning work; extraction errors common |
| Proprietary formats | Vendor lock-in risk | May require paid software or export tools to access |
When to use a meta-search tool vs. a repository directly
Two different strategies depending on how well you know what you're looking for.
| Situation | Best approach |
|---|---|
| You know the exact dataset or repository | Go directly to the source — faster, no noise |
| You know the topic but not the source | Use Google Dataset Search or re3data to discover options |
| You want broad coverage across a field | Use DataONE (environmental) or OpenAlex (publications) for federated search |
| You need data linked to a peer-reviewed paper | Check Dryad or Zenodo — both require DOI linkage |
| You need government statistics | Start at Data.gov (US), Eurostat (EU), or UN Data for international |
Open license cheat sheet
Not all open licenses are equal. Here's what each one means for your project.
| License | Can you use it? | Attribution required? | Commercial use? |
|---|---|---|---|
| CC0 (Public Domain) | Yes, fully | No (but good practice) | Yes |
| CC BY 4.0 | Yes | Yes — cite the source | Yes |
| CC BY-SA | Yes | Yes — share-alike required | Yes, with same license |
| CC BY-NC | Yes for research | Yes | No commercial use |
| ODC-ODbL | Yes | Yes — derivative databases must be open | Yes |
| No license stated | Unclear — contact the owner | N/A | Assume no |