The motors’ fluxlinkage, current and angle obtained from the system with sensors were chosen as the sample data, and the predictive model of rotor position based on relevance vector machine was built by training these sample data. In order to improve the fitting precision and generalization ability of the predictive model, the kernel function parameter in relevance vector machine was optimized by the particle swarm algorithm. By simulation on the test motor, it was verified that the proposed predictive model could estimate the rotor position accurately in the simulation condition and had satisfactory estimation precision.