In this paper, we address discrimination between normal and abnormal left ventricular shapes by capturing deviations from the normal appearance through a new parametric distorted elliptic shape model. To apply the parametric description, we automatically locate the left ventricular region in 4-chamber views and extract its bounding contours and pose. The parametric description of the elliptic fit with minimum alignment error with the bounding contour then becomes the shape descriptor for the bounding contour. Labeled vectors from normal and damaged left ventricular regions are separated into two classes using a support vector machine. Results are presented on a large database of normal and abnormal left ventricular images showing the effectiveness of the parametric features for normal/abnormal discrimination.