Objective Paired associative stimulation (PAS) is a TMS paradigm used to induce long-term potentiation in the human cortex. Little is known about the within-subject consistency of PAS-induced effects. We determined PAS-induced effects and their consistency in healthy volunteers between two PAS sessions. Additionally, we assessed the benefit of applying linear mixed models (LMMs) to PAS data. Methods Thirty-eight healthy volunteers underwent two identical PAS sessions with a >1 week interval. During each session, motor evoked potentials (MEPs) were assessed once before PAS induction and 3 times after at 30 min intervals. Results We did not detect any significant potentiation of MEP size after PAS induction. However, MEP size during PAS induction showed significant potentiation over time in both sessions (LR(1) = 13.36, p<0.001). Nevertheless, there was poor within-subject consistency of PAS-induced effects both during (ICC=0.15) and after induction (ICC=0.03-0.08). Additionally, statistical model selection procedures demonstrate that LMMs are more appropriate than conventional longitudinal models for estimating PAS-induced effects (LR(34) = 214.73, p < 0.001). Conclusion PAS-induced effects are more pronounced during than after induction, have a low within-subject consistency in any phase of the measurement, and are best estimated with LMMs. The implication of our study is that PAS is an inappropriate method to assess the evolution of brain plasticity over time periods longer than the PAS measurement itself.