Abstract: This paper systematically introduces particle swarm optimization algorithm and applies it to the training of neural networks.It advances an improved particle swarm optimization algorithm-Best Influential PSO（BIPSO）,which regards multiple fitness degree（SE and EU） as the guildline of BIPSO and compares it with BP,GA and PSO algorithms.The result of experiment indicates that BIPSO has better performance than other algorithms,finds global best solution more easily and has better astringency.