A double fish-swarm algorithm is proposed to solve reactive power optimization problems in power systems. In this paper, ferocious fish-swarm with the foraging behavior is added in the basic artificial fish-swarm algorithm. The behavior and optimizing process of fish-swam are redefined. The escape-factor is introduced to expand the search space. For enhancing the global optimization ability of the algorithm, the dynamic step is used in the algorithm, which is changing with a try number. The objective function aims at the minimum of the active power loss and voltage deviation, and the reactive power optimization model is established by using the α method and penalty function. In the reactive power optimization objective function, the dynamic penalty coefficient is introduced to improve the application ability of the algorithm. Finally, the basic artificial fish-swarm algorithm, the improved genetic algorithm, and the proposed algorithm have been applied to IEEE 14-bus system for testing the effectiveness. The simulation results show that double fish-swarm algorithm has the best optimization effect in terms of computational accuracy and convergence stability, which is much more in line with the actual operation of the power system.