A modified AFSA (artificial fish swarm algorithm) with parameters dynamic adjusting was proposed to solve the problem of standard AFSA algorithm trapped in local optimal solution and low convergence precision. The advanced algorithm was modified with the following strategies: adjusting the parameters dynamically in the visual field and the congestion factor to improve the searching efficiency, modifying the removing crossover operator to find the crossing point and eliminate it, and designing a further optimizing operator to find the current optimal value again in the current path. The improved AFSA was applied to the TSP problem, and the experiment result has shown that the proposed algorithm has better convergence effect and can improve search performance.