According to Bayes' rule, the brain can generate an optimal estimate for a sensorimotor task by combining prior information of the task statistics with the likelihood obtained from sensory input (Koerding & Wolpert, 2004). Based on our preliminary EEG study, we hypothesized that the dorsal premotor cortex (PMd) is associated with Bayesian estimation during tactile temporal order judgment (TOJ) across both hands (Miyazai et al. 2006; Nagai et al. 2012). Here, we tested this hypothesis using disruptive transcranial magnetic stimulation (TMS). Twelve subjects performed tactile TOJ. The stimulus onset asynchronies (SOAs) were sampled from the Gaussian distribution, biased to the right hand first (mean = 80 ms, SD = 80 ms; R1st-bias) or to the left hand first (mean = -80 ms, SD = 80 ms; L1st-bias). Each participant completed two sessions (1600 trials/session). Following van der Berg et al. (2010), we applied TMS over the right PMd in one session and over the left PMd in another session. For each session, the TMS and no-TMS conditions were interleaved every 200 trials. Under the TMS condition, TMS was applied in 50% of the trials, with the SOA within -40 to 40 ms. We used a paired pulse (ISI = 10 ms) for TMS. The TMS intensity was 110% of the resting motor threshold. The TMS timing was 80 ms after the first tactile stimulus in each trial. We calculated the difference in the point of subjective simultaneity between the R1st-bias and L1st-bias conditions (ΔPSS) using psychometric functions. Positive ΔPSS values imply that the brain implements Bayesian estimation using biased distributions as prior information. The results supported our hypothesis. TMS of the left PMd did not affect ΔPSS (mean ± SEM: no-TMS, 67.8 ± 17.2 ms; TMS, 82.3 ± 20.7 ms; P = 0.51, paired t-test with Holm correction), but TMS of the right PMd increased ΔPSS (no-TMS, 68.2 ± 17.8 ms; TMS, 91.3 ± 16.2 ms; P = 0.02). Greater ΔPSS values imply an increased dependence on prior information. That is, TMS of the right PMd resulted in an overuse of prior information, suggesting that the right PMd is involved in processing of the likelihood.