BACKGROUND: In patients presenting brain metastases as the first manifestation of a previously undiagnosed primary tumour (UDP) histopathological confirmation of the diagnosis can be obtained by either direct surgical sampling of the brain lesion or paraclinical search for an accessible primary tumour. The sequence of the diagnostic work-up and the timing of an eventual neurosurgical intervention are a matter of debate and are mainly influenced by the distribution of primary tumours in UDP patients. The aim of this study was to verify the hypothesis that the distribution of primary tumours differs between UDP patients and the rest of the patients with brain metastases (DP), and to propose a diagnostic work-up specifically tailored to the UDP population. METHODS: Retrospective study on 342 patients admitted to the Lausanne University hospital between 1983 and 1998 with the diagnosis of cerebral metastases. FINDINGS: UDP patients represented 36% of the whole group. Primary tumour location was significantly different between the two groups (p=0.001). Although the lung was the most frequent primary tumour location in both groups (UDP: 60%, DP: 43%), in UDP 14% only of the primaries were found outside of the lung and as much as 26% remained unknown despite thorough investigations. CONCLUSIONS: Our study confirmed the hypothesis that the relative frequency of primary tumours differs between DP and UDP patients. This difference therefore mandates a diagnostic strategy specifically tailored for UDP patients: if a radiological lung investigation clearly remains the best initial step in the work-up of these patients, extensive paraclinical investigations without a clear clinical suspicion should probably not be undertaken if this first survey fails to disclose the primary tumour as only 14% of the patients will actually benefit from it. In this situation, a neurosurgical procedure should probably be considered the most appropriate next step to be taken in order to provide a definitive diagnosis without unnecessary delays.