Background Response rates to repetitive transcranial magnetic stimulation (rTMS) for depression are 25-45%. Biomarkers predicting response to rTMS may reduce treatment burden. TMS-evoked neural activity recorded via electroencephalography (EEG) has potential as a biomarker of treatment response. We examined whether these measures could differentiate responders and non-responders to rTMS for depression. Methods Thirty-nine patients with treatment-resistant major depressive disorder (MDD) and 21 healthy controls received TMS during EEG recordings (TMS-EEG). MDD participants then completed 5-8 weeks of rTMS treatment. Repeated measures ANOVAs compared N100 amplitude, N100 slope, and theta power across 3 groups (responders, non-responders and controls), 2 hemispheres (left, F3, and right, F4), and 2 stimulation types (single pulse and paired pulses with a 100ms inter-pulse interval [pp100]). Results Neither N100 amplitude nor theta power differed between responders and non-responders. The control group showed more negative N100 amplitudes than the combined depression group. Responders showed a steeper negative N100 slope for single pulses and steeper positive slope for pp100 pulses at F3 than non-responders. Exploratory analyses suggested this may have been due to the responder group showing larger late P60 and N100 amplitudes. Receiver Operator Characteristic (ROC) curve analysis indicated the difference between single and pp100 slopes provided excellent sensitivity (1.00), but poor specificity (0.455). Limitations Our study had a small sample size. Conclusion Left hemisphere TEPs, in particular N100 slope, may be useful in predicting non-responders to rTMS treatment for depression. Non-response prediction may be useful in saving clinical resources and patient time.