Cardiac ultrasound can be used to estimate thepatient specific geometry of the heart, e.g., to model theinverse ECG problem. However, segmentation of the surfaceis complicated by artifacts, e.g., speckle noise, and segmen-tation methods are typically validated against manuallydrawn contours. This requires considerable expertknowledge and is time consuming. Hence, wehave devel-oped a setup that allows studying the accuracy of imagesegmentation from cardiac ultrasound. Using a three-dimensional mold constructed from a CT-data set, we havestudied whether the actual geometry can be reconstructedfrom ultrasound images. We illustrate the use of our ap-proach in quantifying the segmentation result for a three-dimensional region-based active contour algorithm.