The neural network model based on Hooke-Jeeves Particle Swarm Optimization(HJPSO) is proposed in this paper for overcoming some shortcomings of BP neural network that is slow at the convergence rate and easy to trap in local minimum.This improved model has utilized prediction of gear heat treatment.The experimental data shows the effect of the improved BP neural model is much better than standard BP neural network in term of the predicting ability and stability.