Combining automatic processing with interactive techniques is proving to be an effective strategy for segmenting complex three-dimensional (3D) medical images. We describe a general 3D image segmentation strategy that draws upon morphological watershed analysis and operator-defined topological cues. Watershed analysis segments a gray scale image into different regions by interpreting the image as a topographic surface. Using readily available interactive techniques, a human operator can easily define cues that specify spatial relationships between regions of interest. Cues defined in such a manner greatly assist subsequent watershed analysis. Results using 3D cardiac images show that this method leads to rapid robust image segmentation.