25 The ability to interpret transcanial magnetic stimulation (TMS) evoked EEG potentials (TEPs) is limited 26 by artifacts, such as auditory evoked responses produced by the TMS coil. TEPs generated from direct 27 cortical stimulation should vary in spatial distribution with stimulation site and from responses to sham 28 stimulation. Responses that do not show these effects are likely to be artifactual. In 20 healthy volunteers, 29 we delivered active and sham TMS to right prefrontal, left primary motor, and left posterior parietal 30 cortex and compared the waveform similarity of TEPs between stimulation sites and active and sham 31 TMS with a cosine similarity-based analysis method. We looked for epochs after the stimulus when the 32 spatial pattern of TMS-evoked activation had greater than random similarity between stimulation sites 33 and sham vs. active TMS, indicating presence of a dominant artifact, such as the auditory brain response. 34 We calculated and binarized the derivatives of the TEPs recorded from 30 EEG channels and calculated 35 cosine similarity between conditions at each time point with millisecond resolution. Only TEP 36 components occurring before approximately 80 ms differed across stimulation sites and between active 37 and sham, indicating site and condition-specific responses. TEP components before about 80 ms can be 38 safely interpreted as stimulation location-specific responses to TMS, but components beyond this latency 39 should be interpreted with caution due to high similarity in their spatial distribution. 3 40 Introduction 41 Transcranial magnetic stimulation (TMS)-evoked EEG potentials (TEPs) are a rich source of 42 neurophysiological data, and can be used to study cortical excitability and effective connectivity [1]. 43 TEPs are obtained by activating the cortex with TMS and measuring the response with 44 electroencephalography (EEG) across the scalp. TEPs have been used to study a variety of phenomena, 45 including consciousness [2,3], memory [4], and the pathophysiology of neurological disorders [5-7]. 46 A major factor limiting the interpretation of the EEG response to TMS is the presence of time-47 locked artifacts. Artifacts from sources outside of the brain, such as electrode noise and recharge artifacts 130 we removed one channel per participant for each stimulation condition (minimum = 0; maximum = 3). 131 We baseline-corrected each epoch by whole-epoch de-meaning and then performed two rounds of ICA to 132 detect and remove artifacts. Both rounds used the FastICA algorithm [23] with a symmetric 133 decomposition approach and hyperbolic tangent contrast function. The first round of ICA included an 134 automated artifact detection algorithm (tesa_compselect) [8] to remove large pulse-related artifacts. The 135 second round was performed to eliminate small artifacts not rejected during the first round. To obtain a 136 consensus about which components should be rejected, authors JR, SH, and MF performed agreement 137 training on a test data set. Each author individually p...