The theory and the algorithm of Particle swarm optimization (PSO) based for neural network training were applied in the research of the modeling of milling force in high speed machining. The PSO was used to train connection weights of multi-layer feed forward neural network until the fitness error tended to be stable. Then BP algorithm was adopted to accomplish cutting force forecasting based on optimized initial weights, which take full use of the global optimization of PSO and local accurate searching of BP. The results of simulation showed that with comparison with other BP algorithms, PSO-BP not only effectively shortens the time of training networks, but also greatly improves the accuracy of prediction and universal approximation. PSO technique can act as an alternative training algorithm for ANNs.