Rejecting bad epochs

Why do I need this step?

The simple fix: reject any epoch in which the amplitude crosses a set limit in any channel.

But rejection has a cost. Every epoch you reject is one trial less in the average, and an average of fewer trials is noisier. So the goal is not to reject everything that looks odd, but to reject what actually harms the average.

How do I do it in EEGLAB?

Start with the dataset D1 epochs bl open, as in “Starting point”.

First: mark the extreme epochs and reject them with the script

The epoch rejection menu is hidden by default in EEGLAB 2026.1, so we start with the script.

1. Copy the whole block from “The equivalent script” below into the command window.
2. The command window prints the number of marked epochs, for example Marked 1 of 80 epochs. The script removes the marked epochs, names the dataset D1 epochs clean, saves it as D1_epochs_clean.set, and adds it to the Datasets menu. The lessons in stations 6 and 7 start from it.
3. If it prints Marked 0 of 80 epochs, there is nothing to reject. The new dataset is a copy of the previous one with the new name, and that is correct.

Second (optional): the same step from the menu

This is an alternative to the script, not an extra step. The menu is Tools › then Reject data epochs › then Reject extreme values, but it is hidden by default: under the Tools menu you will see a gray line that reads (Expand tool choices via “File > Preferences”). To show it:

1. Select D1 epochs bl from the Datasets menu, and save it with File › then Save current dataset as, as D1_epochs_bl.set.
2. Open File › then Preferences, check If set, show all menu items from previous EEGLAB versions. You must restart EEGLAB for this to take effect., then press Ok.
3. Restart EEGLAB: type eeglab in the command window, then open the file with File › then Load existing dataset.

This option stays saved, and it changes how all the menus look compared with the pictures in these lessons, so turn it off after this step. Once the menu shows:

1. Open Tools › then Reject data epochs › then Reject extreme values.
2. The Rejection abnormal elec. values — pop_eegthresh() window opens. Keep Electrode (indices(s), Ex: 2 4 5): at 1:32, that is, all channels.
3. In Minimum rejection threshold(s) (uV, Ex:-20 -10 -15):, type -100 instead of the default -10.
4. In Maximum rejection threshold(s) (uV, Ex: 20 10 15):, type 100 instead of the default 10.
5. Keep Start time limit(s) (seconds, Ex -0.1 0.3): and End time limit(s) (seconds, Ex 0.2): at their values, that is, the whole epoch.
6. Check Reject marked trial(s) and press Ok.
7. A window showing the epochs opens, with the marked epochs highlighted. Review them by eye: you can click an epoch to unmark or mark it.
   • If the command window prints 0/80 trials marked for rejection, there is nothing to reject, and the REJECT button will do nothing. Close the viewing window, open Edit › then Dataset info, type D1 epochs clean in Dataset name, press Ok, and go to step 10.
8. Press REJECT, then confirm with YES.
9. In the new-dataset window, type D1 epochs clean in Name it:, then press Ok.
10. Save it with File › then Save current dataset as, as D1_epochs_clean.set. The lessons in stations 6 and 7 start from it.

How do I write it as code?

% Mark epochs where any channel goes beyond +/-100 uV anywhere in the epoch
EEG = pop_eegthresh(EEG, 1, 1:EEG.nbchan, -100, 100, EEG.xmin, EEG.xmax, 0, 0);
badEpochs = find(EEG.reject.rejthresh);
fprintf('Marked %d of %d epochs\n', numel(badEpochs), EEG.trials);

% Remove the marked epochs
if ~isempty(badEpochs)
    EEG = pop_rejepoch(EEG, EEG.reject.rejthresh, 0);
end
EEG.setname = 'D1 epochs clean';
EEG = eeg_checkset(EEG);
EEG = pop_saveset(EEG, 'filename', 'D1_epochs_clean.set', 'filepath', pwd);

% Add the clean dataset to the EEGLAB window
[ALLEEG, EEG, CURRENTSET] = eeg_store(ALLEEG, EEG);
eeglab redraw

The last argument of pop_eegthresh is 0, meaning “mark only, do not reject”. The rejection happens on the next lines, so you can print the count first.

Which value should I choose?

Rule: the same threshold for every participant and every condition, decided before you see the results. Report the number of rejected epochs per condition in your paper.

How do I check it worked?

• The command window prints the number marked, such as 1/80 trials marked for rejection. In D1 after ICA cleaning, expect a small number, about 0 to 3. It varies slightly from run to run because the ICA result varies.
• Epochs in the main window equals 80 minus the number rejected.
• max(abs(EEG.data(:))) is below 100.
• The file D1_epochs_clean.set is in your working folder.

What mistakes do people make?

• A threshold that is too strict. It throws away good data, so the average comes from few trials and is noisier. The best threshold is not the smallest possible number.
• Rejecting before baseline correction. An epoch shifted away from zero as a whole may cross the limit with no artifact. Correct the baseline first, as in our path.
• Rejecting instead of using ICA. If you reject every epoch with a blink without ICA, you may lose a large part of your data, as the interactive demo shows at ±50. ICA fixes blinks; rejection is for rare problems.
• Very different counts between conditions. If 40% of one condition and 5% of another are rejected, the comparison is not fair. Check the count for each condition.
• Forgetting to save after rejection. In the GUI, rejection creates a new dataset in memory only. Save it before you close EEGLAB.

How do I write it in my paper?

A template: edit the data, settings and versions to match what you ran.

Epochs in which the voltage at any channel exceeded ±100 µV were rejected (pop_eegthresh, EEGLAB 2026.1), leaving [N] of 80 epochs.

Going further

• EEGLAB Tutorial: https://eeglab.org/tutorials/misc/Rejecting_Artifacts_Legacy_Menus.html (Legacy rejection: Rejecting extreme values)
• Makoto's preprocessing pipeline: https://eeglab.ucsd.edu/wiki/Makoto%27s_preprocessing_pipeline (§11 Reject epochs)
• Applied Event-Related Potential Data Analysis (Luck): https://socialsci.libretexts.org/Bookshelves/Psychology/Biological_Psychology/Applied_Event-Related_Potential_Data_Analysis_(Luck) (Chapter 8: Artifact Detection and Rejection)