Observation contributes to motor learning. It was recently demonstrated that the observation of both a novice and an expert model (mixed observation) resulted in better learning of a complex spatiotemporal task than the observation of either a novice or an expert model. In experiment 1, we aimed to determine whether mixed observation better promotes learning due to the information that can be gained from two models who exhibit different skill levels or simply because multiple models, regardless of their level of expertise, better promote learning than would a single model. The results revealed that the observation of both an expert and a novice model resulted in better short-term retention than the observation of either two novice or two expert models. In experiment 2, we wanted to determine whether these benefits would last longer if physical practice trials were interspersed with observation. Mixed and (to some extent) expert observations resulted in better long-term retention than observation of a novice model. We suggest that alternating mixed/expert observation with physical practice trials makes one's error more salient than when all observation trials are completed before one first starts performing the experimental task, which increases activation of the action observation network.