Motor practice is associated with the formation of elementary motor memories. Here we tested in human subjects the hypothesis that observation of a motor training associated with physical practice will modulate the encoding process of a motor memory relative to physical practice alone. Voluntary thumb motions were practiced (i) alone in a direction opposite to the baseline direction of transcranial magnetic stimulation (TMS)-evoked movements (physical practice, PP) and in combination with observation of synchronous movements that were either (ii) directionally congruent (same direction, PP + AOc) or (iii) non-congruent (opposite direction, PP + AOnc) to the practiced ones. We evaluated the following measures of motor memory formation: percentage of TMS-evoked thumb movements falling in the direction of practiced motions, acceleration of TMS-evoked movements along the principal movement axis and corticomuscular excitability of training muscles as indexed by motor-evoked potential amplitudes. Both PP and PP + AOc, but not PP + AOnc, significantly increased the percentage of TMS-evoked movements falling in the practiced direction, changed the compound acceleration vector into the trained direction and enhanced the motor-evoked potential amplitudes in the training agonist muscle. The percentage of TMS-evoked movements falling in the practiced direction increased significantly more after PP + AOc than after PP. Across all measures of motor memory formation, PP + AOc was most efficacious, followed by PP and PP + AOnc. Action observation modulates practice effects on formation of a motor memory. Strengthening of the process of motor memory encoding depends on the directional congruency of the observed model.