Sample recordings · EEG for Beginners

These are real recordings from consumer and research-grade headsets, exported as EDF. They were picked because each one shows something specific: a clean alpha rhythm, line noise, blinks, or the same recording before and after cleaning. Nothing here is a textbook-perfect lab recording, and that is deliberate. This is what EEG looks like when you record it.

All files are anonymised: the patient and recording fields in each EDF header have been blanked, and the start date is set to 01.01.85, the EDF convention for "anonymised". The signals themselves are unchanged.

Loading a sample #

From Python, with the course helpers on your path (run from the code/ folder or add it to sys.path):

python

from eeg101 import load_sample

raw = load_sample("rest_ec")      # short name, or the file name
raw.plot(duration=10)

load_sample reads the file, marks every channel as EEG, renames the four 10-10 temporal sites to their older 10-20 names (T7→T3, T8→T4, P7→T5, P8→T6) so all files match, and attaches template 10-20 electrode positions.

Without the helpers, plain MNE works too:

python

import mne
raw = mne.io.read_raw_edf("data/samples/s01_rest_eyes-closed_19ch.edf", preload=True)

The catalogue #

File	Short name	Channels	Rate (Hz)	Length (s)	Size
`s01_rest_eyes-closed_19ch.edf`	`rest_ec`	19	250	310	2.9 MB
`s02_rest_eyes-open_19ch.edf`	`rest_eo`	19	249	308	2.9 MB
`s03_linenoise_raw_19ch.edf`	`linenoise_raw`	19	250	410	3.9 MB
`s04_linenoise_cleaned_19ch.edf`	`linenoise_cleaned`	19	250	410	3.8 MB
`s05_artifacts_eyes-closed_19ch.edf`	`artifacts_ec`	19	250	73	0.7 MB
`s06_pipeline_stage0_raw_19ch.edf`	`pipeline_raw`	19	250	75	0.7 MB
`s07_pipeline_stage1_rejection_19ch.edf`	`pipeline_rejection`	19	250	46	0.4 MB
`s08_pipeline_stage2_full_19ch.edf`	`pipeline_full`	19	250	74	0.7 MB
`s09_segments_original_8ch.edf`	`segments_original`	8	251	122	0.5 MB
`s10_segments_removed_8ch.edf`	`segments_removed`	8	251	98	0.4 MB
`s11_4ch_eyes-closed.edf`	`headband_ec`	4	250	52	0.1 MB
`s12_4ch_eyes-open.edf`	`headband_eo`	4	250	59	0.1 MB
`s13_4ch_eyes-open_task.edf`	`headband_task`	4	250	48	0.1 MB

The 19-channel files use the standard clinical 10-20 set: Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, O2. The 8-channel files have Cz, F3, F4, O1, O2, Pz, T3, T4; the 4-channel headband has F3, F4, Cz, O1.

Resting state pair (s01, s02) #

The same adult, same session, five minutes eyes closed then five minutes eyes open, on a 19-channel dry-electrode headset. This is the pair to start with. Eyes closed shows a strong 10 Hz posterior alpha rhythm; at O1 and O2 alpha power is about 18 times higher with eyes closed than open. Used in Modules 02, 03, 04, 06, 08, 10 and 13.

Line noise, before and after (s03, s04) #

One eyes-closed recording made somewhere with heavy mains interference. The raw file (s03) has a 60 Hz peak about 60 dB (a million times in power) above the surrounding spectrum, and a second, smaller peak at 70 Hz that is not mains (most likely the hardware itself). s04 is the same 410 seconds after the platform's cleaning pipeline (1 Hz high-pass, 45 Hz low-pass, 60 Hz notch, artifact repair). A residual 60 Hz peak survives, which is a lesson in its own right. Used in Module 07.

Eye artifacts (s05) #

Eyes closed, but the eyes keep rolling: large slow waves at Fp1, Fp2, F7 and F8 while alpha carries on at the back. The last four seconds are saturated junk where the headset came off; find it before you analyse anything. Used in Modules 05 and 09.

Pipeline stages (s06, s07, s08) #

Three exports of the same 75-second recording from the Divergence artifact pipeline:

s06: the raw export. Contains a series of classic sharp blinks, which makes it the best file for practising ICA. Also ends with a few seconds of junk.
s07: rejection only. Bad segments have been cut out, leaving 46 seconds.
s08: the full pipeline. Segments are repaired rather than cut (ASR/ICA-style correction), so the length is back to 74 seconds.

Comparing the three is the clearest way to see what "cleaning" actually does. Used in Module 09 and Lab 2.

Segment rejection (s09, s10) #

An 8-channel headset recording (122 s) and the same file after automatic rejection cut 15 noisy segments (98 s). s10 stores each cut as an annotation such as Noise removed 2.40 s at 00:00:51.823, so you can see exactly what went. The original also has strong slow drift. Used in Modules 05 and 09.

Four-channel headband (s11, s12, s13) #

A low-cost 4-channel headband, about a minute each of eyes closed, eyes open, and eyes open during a mental task. Even with four channels, the eyes-closed alpha peak at O1 is obvious: alpha is about 38% of 1-40 Hz power with eyes closed against about 8% with eyes open. Used in Module 10 and the neurofeedback module.

Things these files will teach you whether you like it or not #

Sampling rates are not always round. Some files say 249 or 251 Hz. Always read raw.info["sfreq"] instead of assuming 250.
Raw files carry a large DC offset. Unfiltered values sit tens of thousands of microvolts from zero. That is the electrode, not the brain. A high-pass filter removes it.
Recordings end badly. People take headsets off before the recording stops. Look at the last few seconds before computing anything.
Old and new electrode names coexist. T3/T4/T5/T6 in one file, T7/T8/P7/P8 in another. Same places.

Public datasets #

For more data, see the resource library. The PhysioNet EEG Motor Movement/Imagery dataset loads directly through mne.datasets.eegbci and is what Lab 1's optional extension uses.