In this paper, a computation model of Available Transfer Capability (ATC) based on Optimal Power Flows (OPF) is established. Improved Artificial Fish Swarm Algorithm (IAFSA) with the advantages of distributed parallel searching ability, strong robustness, good global astringency and easy implement etc is employed to solve this model. Based on the search characteristics of fish swarm, a strategy of non-stationary multi-stage assignment penalty function is proposed to handle inequality constraints during the calculating process. Meanwhile, self-adaptive step, the survival strategy and competition strategy are introduced, which elevate the adaptability of fish swarm and accelerate the speed of convergence of Artificial Fish Swarm Algorithm (AFSA) all the more. The presented method is studied on IEEE-30 bus system. Compared with results gained by Benders decomposition and Improved Particle Swarm Optimization (IPSO), the results demonstrate the feasibility and the validity of this proposed algorithm.