This paper proposes a 3D model retrieval method based on PCA and Hausdorff distance. The geometric center of the root mean square distance (RMS) is calculated to reduce the effect of size difference on similarity analysis. The characteristics of the covariance matrix of the point cloud model is decomposed, and the principal component vector is extracted to construct the local coordinate system. The orthogonal transformation is used to accurately align the main axis of the model, and the optimal rotation matrix is calculated to align the point cloud, so as to eliminate the matching bias caused by the rotation. The spatial index speed is increased by k-d tree, and the bidirectional maximum minimum distance pole between models is calculated simultaneously, using the maximum pole of both as the shape similarity criterion. This method achieves 92.5% matching accuracy on the ModelNet dataset, with 22% less time-consuming than the original Hausdorff distance retrieval, and has good retrieval performance.