High-precision assembly is a very important part in the process of industrial manufacturing. For high-precision components, virtual simulation before assembly is necessary. With the improvement of assembly device accuracy, how to extract surface feature information from device point cloud data accurately and quickly becomes a key problem in virtual assembly. An improved random sample consensus (RANSAC) point cloud segmentation algorithm is proposed to solve the problem of low accuracy and slow speed in extracting surface features of assembly devices due to the large number of iterations and poor robustness of current point cloud segmentation algorithms. Firstly, the device point cloud data is preprocessed to increase the proportion of interior points in the total data set and reduce the number of iterations. Then, the improved RANSAC algorithm is used to calculate the pre-processed data set, estimate the surface model quickly, and extract the geometric features of each surface of the device. Finally, the Euclidean clustering algorithm is used to extract the set of external points, and the error of each surface of the device is calculated based on the surface model, so as to judge whether the assembly of the devices is successful or not. The experimental results show that the algorithm can accurately segment the device surface and find the uneven area, which significantly improves the response speed of the virtual assembly system.