This review analyzes the impact of 2-[(18)F]fluoro-2-deoxy-D-glucose (FDG)-positron emission tomography (PET) in the diagnostic work-up of classic fever of unknown origin (FUO) according to the criteria first proposed by Petersorf in 1961 and later modified by Durack et al. in 1991. Algorithms currently used in this diagnostic process are not strictly evidence based up to now. FDG accumulates in malignant tissues, but also in inflammatory cells by the overexpression of facultative glucose transporter-isotypes (mainly GLUT-1 and GLUT-3) and by an overproduction of glycolytic enzymes. Therefore, this technique covers a broad spectrum of possible etiologies for FUO. Once imaged, these lesions can be further investigated by other (e.g. invasive) and more specific methods. Until now, four prospective studies using FDG-PET in patients with classic FUO, encompassing 167 patients in total are published. Three retrospective studies with 125 patients are also available. These studies are discussed and weighted according to the control of selection-bias that was performed. An interstudy-bias may also be present resulting from a considerable variability in causes of FUO. A low number of diagnostic scans in a study may sometimes be related to a high rate of fevers caused by miscellaneous disorders or to a high rate of undiagnosed patients. In these disease categories, focal pathologies that can be imaged with FDG-PET, are rare. A high number of diagnostic scans is always related to a high prevalence of patients with medium- and large-vessel vasculitis. Available data indicate that FDG-PET has the potential to play an important role as a second line procedure in the management of about 1/3 of patients with classic FUO. It is expected that hybrid imaging (PET/computed tomography [CT]; PET/magnetic resonance imaging [MRI]) will improve the diagnostic impact of FDG-PET further, but prospective data about the value of this methods are currently not available. The question as to how these new techniques can be implemented into an evidence based diagnostic algorithm, can only be resolved within a multidisciplinary setting, avoiding both selection- and interstudy-bias whenever possible.