Parametric shape representations of endocardial contours, obtained with principal component analysis (PCA) and the orthomax criterion, provide compact descriptors for classifying segmental left ventricular wall motion. Endocardial contours were delineated in the left ventricular echocardiograms of 129 patients. Parametric models of these shapes were built with PCA and subsequently rotated using the orthomax criterion, producing models with local variations. Shape parameters of this localized model were used to predict the presence of wall motion abnormalities, as determined by expert visual wall motion scoring. Best results were obtained using the varimax criterion and full variance models. Although traditional PCA models needed 8.0 +/- 3.0 parameters to classify segmental wall motion, only 5.1 +/- 3.2 parameters were needed using the orthomax rotated models (P < .05) to achieve similar classification accuracy. The classification space was also better behaved. Orthomax rotation generates more local parameters, which are successful in reducing the complexity of wall motion classification. Because pathologies are typically spatially localized, many medical applications involving local classification should benefit from orthomax parameterizations.