Purpose: The purpose of this study was to investigate the role of the dorsal premotor cortex (dPMC) in reconstruction and learning of novel finger movement sequence. Especially, different structures of practice (blocked vs. random) were introduced to vary the level of contextual interference, and the dPMC activities during the motor preparation phase were manipulated by the transcranial magnetic stimulation (TMS). Methods: Different patterns of serial key press learning tasks were introduced in random or blocked orders while either the real or the sham TMS was applied at the movement preparation stage. Thus, 56 subjects were randomly assigned into Sham-blocked, TMS-Blocked, Sham-random, and TMS-random practice conditions. 3-way ANOVA were performed to identify the group and condition differences. Results: As a result, the Blocked practice group showed fluent task performances regardless of TMS stimulation during the practice periods, while only the Sham-random practice groups showed the effect of contextual interference in the transfer and retention periods. Conclusion: Significant difference of practice effect depended on the presence of TMS stimulation only with the higher contextual interference in the motor preparation stage. The dPMC seems to be responsible for preparation and reorganization of new motor plans when the variability of the task is high. These results supports the forgetting-reconstruction theory to better explain the effect of contextual interference.