In order to acquire a better extraction of 3D features of building and aimed at the problem that the eight neighborhood search method of extracting 3D features of the building with a sloping roof is not good in precision, in this paper, a smaller threshold elevation difference is set first and then buildings are extracted by using the eight neighborhood search method to improve the accuracy of the algorithm for the extraction of the buildings.The graded differentials in four directions around the point are used to complete the secondary extraction of the feature points cloud to increase the full rate of the extraction of the buildings.The building patch has a large surface area of the chip features and represents the actual number of communicating with the interior point of scene size. Setting of a reasonable threshold can eliminate the vegetation patches and other surface features patches.The results show that the use of the improved method can obviously increase the accuracy and integrity rate of 3D features of building extracted from the mixture of the point cloud and the adaptability of the method in extracting the sloping roof is improved.