The feed system is an important part of the CNC machine tool, and its performance directly affects the machining quality. Establishing an accurate model is the key to achieving feed system servo performance optimization and error compensation. In this paper, a fusion modeling method is proposed for the problems of low accuracy and difficulty in identifying stiffness and damping in the traditional lumped mass modeling method. The proposed method first establishes a rigid model without stiffness and damping, then uses a data-driven model to model the unmodeled dynamics, and finally connects the rigid model and the data-driven model in series. The method simplifies the lumped mass model, avoids the problem that the stiffness and damping are difficult to identify, and the fusion of the data-driven model significantly improves the prediction accuracy of the model. The experimental results show that the fused model has higher prediction accuracy for the frequency response characteristics and displacement of the feed system.