Accurate segmentation of three-dimensional（3D）point cloud on the ice cream board surface is the basis of defect identification and detection for ice cream board. A 3D point cloud segmentation algorithm for ice cream board with various deep defect types is proposed. Firstly，the cross-section data of ice cream board surface is extracted，and the ice cream board types about warp or near-plane are identified by line fitting and the standard deviation criterion. Then，for these two kinds of targets，random sampling consistency（RANSAC）algorithm or iterative nearest point（ICP）registration algorithm are used to achieve point cloud segmentation. The experimental results show that four types of typical targets can be reliably and efficiently segmented by the method presented in this paper. The segmentation accuracy is higher than 94.6%，and the integrity is higher than 91.5%，which lays a foundation for subsequent defect indentification and quantitative detection. The proposed algorithm can effectively improve the segmentation accuracy of wood plate plane and has practical engineering application value.