Medical ultrasound (US) has been widely used for distinguishing benign from malignant peripheral soft tissue tumors. However, diagnosis by US is subjective and depends on the experience of the radiologists. The rarity of peripheral soft tissue tumors can make them easily neglected and this frequently leads to delayed diagnosis, which results in a much higher death rate than with other tumors. In this paper, we developed a computer-aided diagnosis (CAD) system to diagnose peripheral soft tissue masses on US images. We retrospectively evaluated 49 cases of pathologically proven peripheral soft tissue masses (32 benign, 17 malignant). The proposed CAD system includes three main procedures: image pre-processing and region-of-interest (ROI) segmentation, feature extraction and statistics-based discriminant analysis (DA). We developed a depth-normalization factor (DNF) to compensate for the influence of the depth setting on the apparent size of the ROI. After image pre-processing and normalization, five features, namely area (A), boundary transition ratio (T), circularity (C), high intensity spots (H) and uniformity (U), were extracted from the US images. A DA function was then employed to analyze these features. A CAD algorithm was then devised for differentiating benign from malignant masses. The CAD system achieved an accuracy of 87.8%, a sensitivity of 88.2%, a specificity of 87.5%, a positive predictive value (PPV) 78.9% and a negative predictive value (NPV) 93.3%. These results indicate that the CAD system is valuable as a means of providing a second diagnostic opinion when radiologists carry out peripheral soft tissue mass diagnosis.