A parallel cooperative particle swarm optimization（PCPSO） algorithm is proposed to solve the large-scale optimal reactive power dispatch（ORPD） problem， which is highly dimensional， nonlinear， and discontinuous. The PCPSO algorithm is based on message passing interface using two levels of parallelism. The first level consists in decomposing the original optimization problem into several smaller sub-problems， each of which is dealt with by a sub-particle swarm. By coordinating their efforts， the sub-particle swarms jointly look for the optimal solution. The second level of parallelism lies in finding the solution using multiple processes in parallel when solving sub-optimization problems with the particle swarm optimization algorithm， with the computation assigned equally among the processes. In addition， time varying acceleration coefficients are used to enhance the global search ability of the PSO algorithm， and dynamic adjustment has been made of its parameters in the process of optimization. The PCPSO algorithm is implemented on a PC-cluster to solve the ORPD problems of an IEEE 118-bus test system and an IEEE 300-bus test system. Simulation results show that the proposed method can yield high-quality results and significantly reduce the optimization time.