Abstract: Aiming at the problem of poor adaptive ability of the triangular mesh automatic segmentation method, an interactive semi-automatic segmentation method based on quadric surface fitting is proposed. Firstly, based on the framework of variational method, quadratic surface fitting and segmentation on the input mesh are processed. And then user interaction is introduced to guide the segmentation process, the interactive operations can be used interchangeably with iterative optimization algorithm. Specifically, the algorithm includes the following three steps: initialization based on random sample consensus (RANSAC), iterative optimization, interactive merging and segmentation. Among them, the initialization method can obtain a good segmentation result. Then an iterative fitting operation is adopted to optimize the segmentation result. Finally, the interactive operations can be used interchangeably with iterative optimization steps, meanwhile the user’s subjective prior knowledge is used to perform local operation on the part where the segmentation is not accurate enough, so that the segmentation effect is more accurate and authentic. The experimental examples show that compared with other methods, our approach has obvious advantages in convergence speed and fitting error. Further, we can accurately segment the mechanical models with complex structure and noise.