We have developed a method for distinguishing benign from malignant focal liver lesions based on multiscale texture analysis. In this method, ROIs extracted from within the lesions were decomposed into subimages by wavelet packets. Multiscale texture features were calculated from these subimages based upon a single-scale feature defined on the original ROIs. An artificial neural network (ANN) was used for combining these multiscale features for classification of lesions, and its performance was measured by the area under the receiver operating characteristic curve (Az). A subset of the multiscale features that yields the highest performance is selected in a step-wise manner as the wavelet packet decomposition is performed. Three single-scale features, i.e., entropy, root mean square, and first moment of the power spectrum, are used to generate the multiscale texture features. In an analysis of 193 ROIs consisting of 50 hemangiomas (benign lesions), 69 hepatocellular carcinomas, and 74 metastases (both malignant lesions), the multiscale features yielded a high Az value of 0.92 in distinguishing benign from malignant lesions, whereas the single-scale features yielded only 0.70. Our multiscale texture analysis method can effectively differentiate malignant from benign lesions, and thus can increase the accuracy of diagnosis of focal liver lesions in sonography.