Prostate contour extraction is an important stage in automatic processing and three dimensional modeling of prostate images acquired from ultrasound devices. In this study, a new approach is presented for extracting prostate boundary from two dimensional ultrasound images. Firstly, noise in the rectangular subimage containing the prostate is reduced by Gaussian filter. This image region is divided into four subimages, each containing a different prostate part. Binarization by adaptive thresholding and then morphological closing and opening operations are applied to each subimage. Then, the edge points that are candidates to prostate contour are found according to local criteria in different parts of prostate. In selection of edge points, pixels are examined if they have special neighborhoods. The points that are satisfying the conditions and close to each other are combined. By using this proposed approach, the final contour of the prostate can be quite appropriately found at the first iteration. Thus, the result is obtained more quickly than the deformable models, that need to many iteration steps.