A curated list of places to get real neural data, and a few other kinds, with a sentence on what each one is good for and what it takes to get in. Everything on this site’s learning path uses data from this page, and the first section is the file you already have. The list is opinionated: it is the datasets I have used, taught with, or would point a new student at, not everything that exists. Last revised September 2026, with every link checked.
Contents: Used on this site · Spikes and population recordings · EEG, MEG and ECoG · MRI and fMRI · Anatomy, connectomes and cell types · Clinical and health data · Aggregators, formats and tools · Beyond neuroscience
Used on this site
Every lesson on this site runs on real data, and these are the files. If you are new, start here: they are small, documented, and you already have code that reads them.
- Steinmetz 2019, single-session extract (2.4 MB)Open
One session of the Steinmetz Neuropixels recordings, cut into trials and binned at 10 ms: 698 neurons across 12 brain areas, 340 trials, stimulus contrasts and the mouse’s responses. Used in lessons 2, 3 and 4. Loads with one line of NumPy. - Steinmetz 2019, Neuromatch Academy version (all sessions)Open
All 39 sessions from Steinmetz, Zatka-Haas, Carandini and Harris, Nature 2019, packaged by Neuromatch Academy as three NumPy files of 13 sessions each; this link is the first part (88 MB), and the Neuromatch loader notebook fetches all three. Spike counts, brain areas, wheel movements, pupil, licks. The spike-time version used in lesson 1 is at osf.io/sy5xt. - Steinmetz 2019, original releaseOpen
The raw spike times, waveforms and behavioural data as published, in the Open Neurophysiology Environment layout. Larger and less tidy than the Neuromatch version, but complete. - Neuromatch Academy project datasetsOpen
Loader notebooks and documentation for every dataset used in the Neuromatch computational neuroscience course: Steinmetz spikes, Stringer’s 10,000-neuron calcium imaging, Kay’s fMRI, the Human Connectome Project task fMRI, ECoG, and more. The best-documented set of starter datasets in the field.
Spikes and population recordings
Extracellular electrophysiology: single neurons and populations recorded with Neuropixels and other probes, mostly in mice, plus the primate motor cortex datasets that brain-computer interface work is built on.
- Allen Brain Observatory: Visual Coding NeuropixelsOpen
58 sessions, about 100,000 neurons, six probes per mouse across visual cortex, thalamus and hippocampus, all watching the same standard set of visual stimuli. The AllenSDK Python package downloads and caches it. The companion two-photon calcium dataset covers 60,000 neurons across cortical layers and cell types. - Allen Brain Observatory: Visual BehaviorOpen
Neuropixels and two-photon recordings from mice performing an image-change detection task, with behaviour, running and pupil. The natural next step after Steinmetz if you want a decision task with more trials per animal. - International Brain Laboratory: brain-wide mapOpen
A standardised decision task recorded in 12 labs with Neuropixels across the whole mouse brain: hundreds of sessions, hundreds of thousands of neurons, one data format, one Python API (ONE). The largest single systems-neuroscience dataset there is. - DANDI ArchiveOpen
The NIH archive for cellular neurophysiology, everything in Neurodata Without Borders (NWB) format: Neuropixels, calcium imaging, patch clamp, human intracranial, from hundreds of labs. Streamable from Python without downloading whole files. If a recent electrophysiology paper shared its data, it is probably here. - Neural Latents BenchmarkOpen
Four curated datasets of monkey motor, premotor, somatosensory and frontal cortex during reaching and timing tasks, with a standard evaluation for latent-variable and decoding models. Hosted on DANDI. Where to go for the primate reaching data that most BCI decoding papers use. - CRCNS: Collaborative Research in Computational NeuroscienceRegistration
The original neuroscience data-sharing site, going back to 2008: retina, visual cortex, hippocampus (the classic hc-2 to hc-11 place-cell recordings), auditory cortex, motor cortex, and more. Formats vary by dataset; free registration. - Stringer 2019: spontaneous activity in 10,000 neuronsOpen
Two-photon calcium imaging of about 10,000 neurons in mouse visual cortex at once, with the face video that turned out to explain much of the activity. A standard dataset for dimensionality and behaviour-related analyses; also available in a small Neuromatch version. - MICrONS functional dataOpen
Calcium imaging of about 75,000 neurons in a cubic millimetre of mouse visual cortex, with the electron-microscopy reconstruction of the same tissue. The first dataset where you can go from a neuron’s responses to its synapses.
EEG, MEG and ECoG
Scalp and intracranial recordings from humans. This is where most brain-computer interface and clinical work starts, and where the data is easiest to collect but hardest to interpret.
- PhysioNet: EEG Motor Movement/ImageryOpen
109 subjects, 64-channel EEG, real and imagined hand and foot movements. The most used open motor-imagery dataset and the standard first BCI project. - PhysioNet: CHB-MIT scalp EEGOpen
Continuous scalp EEG from 22 paediatric epilepsy patients with 198 annotated seizures. The standard benchmark for seizure detection. - PhysioNet: Sleep-EDF ExpandedOpen
197 whole-night polysomnograms with expert sleep-stage labels. The standard automatic sleep-scoring benchmark. - PhysioNet: all databasesOpen
The full index. Also has ECG, ICU waveforms, gait and many more EEG sets. Everything is open or credentialed, well documented, and readable with thewfdbPython package. - OpenNeuroOpen
Over a thousand BIDS-formatted datasets: fMRI, MRI, EEG, MEG, iEEG and PET, free to download with no registration. The default place to share human neuroimaging data, and the first place to look for it. - BCI Competition datasetsRegistration
The Berlin BCI competitions II to IV (2003 to 2008): motor imagery, P300 spellers, ECoG finger movements. Small and old but still the datasets against which BCI algorithms are compared. Free registration for downloads. - MOABB: Mother of All BCI BenchmarksOpen
A Python package that downloads dozens of public EEG BCI datasets (motor imagery, P300, SSVEP) into a common format and runs standard pipelines on them. The fastest way to compare a method across many datasets. - BNCI Horizon 2020 databaseOpen
The Graz BCI datasets and others from the European BNCI project: motor imagery, error-related potentials, SSVEP, stroke patients. - Temple University Hospital EEG CorpusRegistration
Nearly 27,000 clinical EEG sessions from 15,000 patients, with subsets annotated for seizures, artefacts and abnormality. By far the largest clinical EEG resource. Free, but you sign a data-use agreement. - ERP COREOpen
40 participants, six standard event-related-potential paradigms (N170, MMN, N2pc, N400, P3, LRP/ERN), with scripts. Built for teaching ERP methods. - DEAPRegistration
32 participants watching music videos, 32-channel EEG plus peripheral physiology, rated for arousal and valence. The standard emotion-recognition EEG dataset. The Queen Mary server is intermittently offline; try again later rather than assuming it is gone. - MNE-Python sample datasetsOpen
Small, clean MEG and EEG datasets that download with one function call (mne.datasets.sample.data_path()) and are used throughout the MNE tutorials. The easiest way to get a first MEG file open. - National Sleep Research ResourceApplication
Tens of thousands of polysomnograms from large cohort studies (SHHS, MrOS, MESA and others), with harmonised variables. Application required, but routinely granted for research. - iEEG.orgRegistration
Intracranial EEG from epilepsy patients, human and animal, with tools for annotation and sharing. Free registration.
MRI and fMRI
Structural and functional imaging. Files are large; most of these are best used through their own download tools rather than a browser.
- Human Connectome ProjectRegistration
1,200 healthy young adults: structural, diffusion, resting and task fMRI, and MEG for a subset, at exceptionally high quality. Free after accepting the data-use terms; the Young Adult release is the one most methods papers use. - Natural Scenes DatasetRegistration
Eight participants, each viewing about 10,000 natural images over 30 to 40 sessions of 7T fMRI. The dataset behind most recent work on decoding and reconstructing images from brain activity. - BOLD5000Open
Four participants viewing about 5,000 images from computer-vision datasets (ImageNet, COCO, Scenes), slow event-related fMRI. Designed for comparing brains with deep networks. - ABIDE: Autism Brain Imaging Data ExchangeOpen
Resting-state fMRI and structural MRI from over 2,000 individuals, about half with an autism diagnosis, pooled from 17 sites. Preprocessed versions are available. - OASISRegistration
Open Access Series of Imaging Studies: structural and functional MRI, PET and clinical data across the adult lifespan and Alzheimer’s disease, from Washington University. - IXI datasetOpen
Nearly 600 healthy adults, T1, T2, PD, MRA and diffusion MRI from three London hospitals. Small, clean, and a common starting point for segmentation and age-prediction models. - ADNI: Alzheimer’s Disease Neuroimaging InitiativeApplication
Longitudinal MRI, PET, CSF biomarkers, genetics and cognitive scores from thousands of participants along the Alzheimer’s spectrum. The standard dataset for disease-progression modelling. Application required. - PPMI: Parkinson’s Progression Markers InitiativeApplication
The Parkinson’s counterpart to ADNI: imaging, biological samples, genetics and clinical assessments, longitudinal. Application required. - UK BiobankApplication
Half a million participants with genetics, health records and lifestyle data, and brain MRI for tens of thousands of them, heading for 100,000. The largest population imaging study; access is by approved application and a fee. - NeuroVaultOpen
Not raw data but statistical maps: thousands of unthresholded fMRI results uploaded by their authors. Useful for meta-analysis and for checking a claim against the actual map.
Anatomy, connectomes and cell types
What the brain is made of and how it is wired. Mostly browsable online, with programmatic access for the serious.
- Allen Brain AtlasOpen
Gene expression, connectivity, cell types and reference atlases for mouse and human, with an API and the AllenSDK. The Mouse Brain Connectivity Atlas and the Cell Types database (morphology and electrophysiology of thousands of individual neurons) are the two most used pieces. - MICrONSOpen
A cubic millimetre of mouse visual cortex reconstructed from electron microscopy: about 200,000 cells and 500 million synapses, with the functional imaging of the same neurons. Browsable online; queryable from Python. - FlyWireOpen
The complete connectome of an adult fruit fly brain: about 140,000 neurons and 50 million synapses, proofread and annotated with cell types. The first whole-brain connectome of an animal that can walk, fly and learn. - Janelia hemibrainOpen
Half of a fly’s central brain at synapse resolution, from the FlyEM project, with the neuPrint interface for querying circuits. - H01: a petabyte of human cortexOpen
One cubic millimetre of human temporal cortex imaged with electron microscopy by Harvard and Google: 57,000 cells, 150 million synapses. Browsable in Neuroglancer. - NeuroMorpho.orgOpen
Over 250,000 digitally reconstructed neuron morphologies from hundreds of labs and dozens of species, in a standard format. The dataset for any question about dendritic shape. - ModelDBOpen
Not data but models: thousands of published computational neuroscience models with runnable source code, from single ion channels to networks. If a paper’s model is anywhere, it is here.
Clinical and health data
Records, cohorts and genomics for the precision-medicine side of this site. Nearly all of it requires credentials, because it is about real patients; the applications are routine and worth doing.
- MIMIC-IVRegistration
De-identified electronic health records from about 300,000 patients at a Boston hospital’s ICU and emergency department: vitals, labs, medications, notes, and linked waveforms and chest X-rays in companion datasets. The standard dataset for clinical machine learning. Requires PhysioNet credentialing and a short ethics course. - eICU Collaborative Research DatabaseRegistration
200,000 ICU admissions from over 200 US hospitals, the multi-centre complement to MIMIC. Same credentialing. - n2c2 / i2b2 clinical NLP datasetsRegistration
De-identified clinical notes annotated for the National NLP Clinical Challenges: concept extraction, medication, temporal relations, cohort selection. The benchmark datasets for clinical text. - NIMH Data ArchiveApplication
NIH-funded human subjects data in mental health and development, including the ABCD Study (12,000 children followed from age 9 with imaging and cognition) and the autism collections that used to be called NDAR. Application required. - FITBIR: traumatic brain injuryApplication
The federal repository for TBI research data: imaging, clinical outcomes, biomarkers. Application required. - 1000 Genomes / International Genome Sample ResourceOpen
Whole-genome sequences of 2,504 individuals from 26 populations, fully open. The reference panel for human genetic variation. - gnomADOpen
Aggregated variant frequencies from over 800,000 exomes and genomes. The first place to check whether a variant is common. - NCBI GEO: Gene Expression OmnibusOpen
Over 200,000 gene-expression studies, including a large share of the brain transcriptomics literature, in a searchable archive with a Python and R ecosystem for loading it. - HealthData.gov and CDC WONDEROpen
US public-health statistics and datasets: mortality, hospital quality, disease surveillance. Aggregate rather than patient-level, so no application needed. - European Health Data (ECDC via data.europa.eu)Open
The EU open-data portal, including the European Centre for Disease Prevention and Control’s surveillance datasets.
Aggregators, formats and tools
Where to search when none of the above fits, and the standards that make the data usable.
- Neurodata Without Borders (NWB)Open
The standard file format for neurophysiology, used by DANDI, the Allen Institute and the IBL. Learn to read it withpynwband most modern datasets open the same way. - BIDS: Brain Imaging Data StructureOpen
The standard folder layout for MRI, EEG, MEG and iEEG datasets, used by OpenNeuro. A BIDS dataset can be fed straight into standard preprocessing pipelines such as fMRIPrep and MNE-BIDS. - brainlife.ioRegistration
A free platform that runs standard neuroimaging pipelines in the cloud on public or uploaded data. Useful when a dataset is too large for a laptop. - EBRAINS Knowledge GraphOpen
The European Human Brain Project’s data catalogue: atlases, electrophysiology, imaging, models, from rodents to humans. - DataLadOpen
A version-control system for datasets, built on git and git-annex, with a large catalogue of neuroscience datasets (including all of OpenNeuro) that you can clone and fetch selectively. - Registry of Open Data on AWSOpen
Large public datasets hosted free on Amazon S3, including the Allen Brain Observatory, the IBL, OpenNeuro, MICrONS, the HCP and the 1000 Genomes. Streamable without downloading. - Google Dataset SearchOpen
A search engine over dataset metadata across repositories. Good for finding the odd dataset that lives on a lab website. - Hugging Face datasetsOpen
A growing number of EEG, MEG and neural datasets in a format that loads in one line, including MindBigData (EEG while viewing digits) and several Neuropixels conversions. - meagmohit/EEG-DatasetsOpen
A community-maintained GitHub list of public EEG datasets, grouped by paradigm: motor imagery, ERPs, emotion, clinical, sleep. Longer than this page and updated by many hands. - Awesome Public Datasets: neuroscienceOpen
The neuroscience section of the largest general list of open datasets on GitHub.
Beyond neuroscience
General-purpose datasets from the original version of this page, kept because they are still useful for learning machine learning and for the occasional side project.
- UCI Machine Learning RepositoryOpen
The classic collection of small tabular datasets (Iris, Wine, Adult, and about 600 others), including a few EEG sets. Where most people’s first classifier was trained. - Kaggle datasetsRegistration
Tens of thousands of community-uploaded datasets with notebooks attached, including many neuroscience and medical ones. Quality varies; the notebooks are often more useful than the data. - Enron email corpusOpen
Half a million emails from 150 Enron employees, released during the investigation. The standard corpus for email classification and network analysis. - Data.govOpen
The US federal open-data catalogue, about 300,000 datasets. - Nasdaq Data Link (formerly Quandl)Registration
Financial and economic time series. Some free, most paid. - OpenSecrets bulk dataRegistration
US campaign finance and lobbying data from the Center for Responsive Politics. - Internet ArchiveOpen
Books, web snapshots, audio, video and a long tail of uploaded datasets.
Before you use any of it
Four habits that will save you from the mistakes everyone makes with someone else’s data:
- Read the paper first. The dataset is a by-product of an experiment, and the experiment’s design (what the animal or person was doing, how trials were structured, what was excluded) is in the paper, not the README.
- Check the licence and cite it. Most of the datasets here are CC-BY or similar, which means you can use them for anything if you credit the authors. A few forbid redistribution, which matters if you want to host an extract as this site does.
- Plot the raw data before computing anything. A raster, a trace, a slice. Half of the problems with public data (a dead channel, a mislabelled trial, a unit mix-up) are visible in the first plot and invisible in the summary statistics.
- Hold something out. If you are going to fit a model, decide now which sessions or subjects you will never look at until the end. Lesson 3 is about why.
Know a dataset that belongs here, or a link that has died? Tell me on Twitter at @thinquanaut.