Due to the rapid development of computer, sensor, and automatic control technologies, the amount of data generated during product design and manufacturing is increasing significantly. The product data bank is large, complex, heterogeneous, and often fast-changing; it is difficult to integrate heterogeneous models using the conventional method. Therefore, a semantic feature fusion-based heterogeneous model integration method is proposed. First, the error in the geometric dimensions and position are extracted using model registration. Second, the basic geometric feature is obtained using slippage analysis. Third, the extracted data, such as the basic geometric feature and the error in the geometric dimensions and position, are fused into the design model using the level set method. Finally, the marching cubes method is introduced to reconstruct the surface of the fused model. The empirical results demonstrate that the proposed algorithm can integrate all types of semantic features and geometric features into a basic product model effectively and efficiently.