PET imaging is now considered a gold standard tool in clinical oncology, especially for diagnosis purposes. More recent applications such as therapy follow up or tumor targeting in radiotherapy require a fast, accurate and robust metabolically active tumor volumes on emission images, which cannot be obtained through manual contouring. This clinical need has sprung a large number of methodological developments regarding automatic methods to defined tumor volumes on PET images. This paper reviews most of the methodologies that have been recently proposed and discusses their framework and methodological and/or clinical validation. Perspectives regarding the future work to be done are also suggested.