A retrospective study was carried out to determine the performance of 18F-fluorodeoxyglucose positron emission tomography (FDG-PET) in patients with unknown primary tumours presenting with metastases external to the neck. All patients referred to an academic PET centre (July, 1997 to December, 2000) presenting with an extracervical metastasis and no prior systemic therapy were eligible. The minimum follow-up period was 11 months. From 63 eligible cases, known metastases were FDG avid in all but one neuroendocrine process. PET scans were retrospectively classified as positive for a primary tumour (n=29), i.e. revealing at least one anatomical site suspected to be the primary tumour. This was confirmed in 16, either by histology (n=10) or radiological and clinical follow-up (n=6). There were four false positive cases. In nine patients, the primary tumour was never confirmed. Of the remaining 33 negative PET scans the primary tumour was clinically not found in 18. Follow-up and additional pathology investigations demonstrated the primary tumour in 15. A survey on clinical usefulness of PET (response rate 83%) suggested that PET positively contributed to diagnostic understanding in 29 of 52 evaluable cases. Applied late in the diagnostic trajectory, approximately four patients need to be scanned by PET in order to find one primary tumour. However, in addition to direct demonstration of unknown primaries, there appears to be a positive effect on the diagnostic work-up of these patients of a similar magnitude.