A new multilevel thresholding method is proposed using sample moment function (SMF). For the binary case the method can make use of the a priori information on whether the target object is found on the darker or lighter part of the scene. Extensive comparisons using a database of NDT images show that the proposed method outperforms in competition 40 other thresholding methods from the literature M. Sezgin et al. (2001) (2003), based upon both objective scores and subjective evaluation.