A semi-automatic method for three-dimensional segmentation of medical images is proposed. A multiresolution representation is achieved through the application of morphological filters, which assures causality for image extrema. This allows for a compact scale space representation, in which each extremum is assigned a scale value. Interactive selection of the interesting extrema of the image is carried out, aided by this scale information and other relevant features. Extrema selected are then used as markers in three-dimensional watersheds calculation. The system has been developed and tested under low cost platforms, and can be the base for totally automatic, knowledge based segmentation systems.