Functional volumes determination is a crucial step for several applications in oncology like therapy assessment or image-guided radiotherapy treatment planning. There is currently no consensus about the appropriate method for an automatic definition of the tumoural volume on functional emission images (e.g. 18F-FDG PET), because they are characterized by a large variability of noise, textures and contrasts, as well as shapes and uptakes of tumours. We propose a novel automatic method that was validated for robustness and accuracy on phantom acquisitions, realistic simulations and clinical images of complex tumours. This method outperforms the reference thresholding methodologies and may have an impact in several PET applications in oncology. (C) 2009 Elsevier Masson SAS. All rights reserved.