Removing trials with large saccades reduces the confounding effect of eye movements. Two ways of detecting deviating eye movements are shown. In the first method (a,b) those trials are removed in which the eye was a certain degrees of visual angle (DVA) away from the fixation point. In the second method (c,d) those trials are removed in which at least one saccade is made. Saccades are detected by thresholding the eye movement velocity. By increasing the thresholds, more trials stay in the analysis that might contain systematic eye movements (a,c), which in turn increases classification accuracy in the active condition (b,d). An asterisk denotes a significant difference between the active session and the passive session (P < 0.05, Bonferroni corrected). Solid lines show the session-specific median values and dots represent individual datapoints. The gray dashed line represents chance level decoding (i.e., 12.5 percent). Note, a classification accuracy of −5% denotes that too many trials were removed to be able to perform decoding. Additionally, the untreshholded results are shown at the end marked as ‘none’ in all figures, which are the results from the original analysis as in Fig. 2.