In this paper an automated method for volume contouring in PET images is presented. It is a slice-by-slice approach based on marker-controlled watershed segmentation applied to gradient image. A preprocessing step, based on geodesic transformations, is proposed to get a well defined boundary gradient of the region to be segmented (the tumour in our case), since the watershed results are improved. Moreover, a scheme of marker selection is proposed taking into account a priori segmentation knowledge from previous segmented slices. The method has been validated along with other 32 methods in a wide study  using phantom and real data. Regarding its degree of interactivity, our method obtained the highest accuracy results in patient’s data (
* = 0.694) and similar results (
* = 0.670) in the case of phantom’s data.