Commercial animation software utilizes its crowd feature based on agent technologies. Using an intelligent agent for one character allows animators to easily modify a specific character’s behavior in detail, while most other characters can still use the same behavioral template. An agent based crowd, however, suffers from poor performance because a CPU needs to calculate each and every agent’s decision. This thesis presents Cellular Flocking, an approach for reducing the CPU load. By giving agents in the same map cell a shared brain, a group decision can be made using flocking algorithm at cellular automata level. This reduces the calculations significantly. Maintaining the distance among agents and computing agents’ direction are made into group decisions, while collision avoidance is omitted. The prototype was implemented in Max Script. Results show that the proposed technique not only significantly reduces the calculations, but also maintains acceptable group movement in a limited distance.