To solve the high labor cost and low efficiency problems of complex tank members extraction, we propose a point cloud segmentation algorithm which is applicable to tank of plane bulkhead type. The steps of this approach are as follows: the original point cloud is firstly transformed to the independent coordinate, whose X, Y and Z axes are defined as the longitudinal, transverse and vertical directions of tank, based on seed sets building, normal estimation and linear fitting; then clouds of each tank member are segmented in the best order by plane fitting of random sample consensus (RANSAC) method according to the distribution of complex tank members. Experimental results on two sets of point clouds show that the proposed algorithm can quickly, accurately and automatically segment the clouds of main members from unorganized clouds, which demonstrate the high reliability and practical value of the proposed method.