This study addressed the issue of learning fish locomotion patterns, and was realized within an international broader collaboration: the “Fish locomotion project” which aimed at designing and testing multimedia tools for training on fish species conservation. The issue addressed professionals’ skills involved in fish species surveys. They have to identify fish species from viewing videos recorded in oceans (or lakes). Fish species recognition relies on using conventional biological fish classification based on body shapes features. However, very often, when the viewing conditions become difficult (water turbidity), the conventional shape based recognition appears to be no longer effective. Another classification, based on locomotion types, was developed by biologists. This paper presents the results of a study about the internal animation generation effect. Within the theoretical framework of cognitive imagery, and mirror neurons systems in neuroscience, the present experiment investigated the effect of imagining versus viewing animations of fish locomotion. Novices (students) learned fish locomotion, from 3D animation. During a perceptual learning task, we compared 4 presentation conditions of the fish animation: depending of the group, learners could view either, the head of the fish only, the middle of the body only, the tail only or the whole fish, control condition. Eye tracking was recorded in order to assess “internal animation processing”. Post test was a recognition task of the fish locomotion (versus other fish movement, such as an eel). Results indicated that learning in fish parts only conditions produced better learning gains compared to the control condition. Implication of the results for instructional design in training is discussed.