A multimodal industrial anomaly detection method based on normalizing flow is proposed to address the issue of interference between high-dimensional features in multimodal industrial detection, resulting in unsatisfactory detection rates. Firstly, the depth information of the 3D point cloud of the image is extracted and added to the RGB image as the fourth channel to generate the fused RGBD image. Then, the fused image features are extracted using a pre-trained feature extraction network. Finally, a normalizing flow model for anomaly detection is obtained using feature training. The experimental results show that the anomaly detection model achieves an average Pixel AUROC of 95.8% and an average AUPRO of 86.2% on the MVTec 3D-AD dataset, which is an improvement of 2.6% and 9.1%, respectively, compared to other models.