A range image segmentation algorithm based on Gaussian mixture model of surface normal is proposed. It decreases the times of clustering computing by fully utilizing the physical meaning of Gaussian mixture model of surface normal, and achieves automatic model selection via the posterior probabilities computed from the model parameter estimated by Expectation-Maximization (EM) algorithm. Experimental results on 60 real range images from two kinds of range cameras are compared objectively with some popular segmentation algorithms.