With the aim to improve forecasting accuracy of the normal contact stiffness of machined joints, the modified particle swarm optimizer (MPSO) algorithm was proposed. The BP neural network parameters were optimized by the MPSO algorithm. The normal contact stiffness of machined joints was forecasted under different experimental conditions, and the relative errors were analyzed. The results showed that the forecast precision could reach to 92%, and the contact stiffness of machined joints was modeled for various affecting factors. 