In this paper, a novel algorithm in the contourlet domain is presented for segmenting lumbar vertebra in CT image. Firstly we apply the contourlet transform to the image, use its multiscale and multidirectional characteristics to get the coefficients that they can effectively capture the contours of an image and depict them sparsely. Secondly, C-Means clustering algorithm is applied to derive the edge information from the contourlet coefficients. Experimental results are encouraging, which illustrate the effectiveness of the algorithm. Compared with the wavelet-based method, our algorithm can get a satisfactory results.