This paper proposes a shape-based neurobiological approach for action recognition. Our work is motivated by the successful quantitative model for the organization of the shape pathways in primate visual cortex. In our approach the motion-salience region (MSR) is firstly extracted from the sequential silhouettes of an action. Then, the MSR is represented by simulating the static object representation in the ventral stream of primate visual cortex. Finally, a linear multi-class classifier is used to classify the action. Experiments on publicly available action datasets demonstrate the proposed approach is robust to partial occlusion and deformation of actors and has lower computational cost than the neurobiological models that simulate the motion representation in primate dorsal stream.