In order to complement the missing point cloud and improve the selection of single-view mirror symmetry planes, and solve the problem of low parameter dimensions in the existing cow weight estimation model, a method of dairy cow weight estimation based on 3D reconstruction was proposed. Firstly, a cow depth video acquisition platform was built, and the cow’s top and side perspective data were collected by using the Kinect camera. After selecting the synchronized top and side view frames in the depth video, they were converted to point clouds, and the complex background was removed to extract the cow points cloud. Then it was proposed to use the side view point cloud of different frames to complete the missing part of the selected side view point cloud, and after registering the top view and side view point clouds, for the single view side view point cloud, a method was proposed to select the symmetry plane based on the position of the cow spine in the overlook point cloud, so the dual-view side-view point cloud was obtained, and the reconstruction of the point cloud on the surface of the cow was completed. Finally, point cloud surface reconstruction was carried out, and the volume and surface area of the surface model were used to establish a cow weight estimation model. The data of 29 cows were used to verify the model, and the results showed that the surface area of the cow’s curved model, the volume of the curved model, excluding the limbs and head, and body weight were significantly positively correlated. The absolute error of weight estimation was between -18.67kg and 23.34kg, the relative error was less than 3.40%, and the average relative error was 2.04%.