The purpose of this study was to calculate disease probabilities based on data of patients with head and neck cancer in the register of our institution and to perform a systematic review of the available data on the accuracy of PET in the primary assessment and follow-up of patients with head and neck cancer. The pre-test probability of head and neck cancer among patients in our institutional data registry was assessed. Then the published literature was selected and appraised according to a standard protocol of systematic reviews. Two reviewers independently selected and extracted data on study characteristics, quality and accuracy. Accuracy data were used to form 2 x 2 contingency tables and were pooled to produce summary receiver operating characteristic (ROC) curves and summary likelihood ratios for positive and negative testing. Finally post-test probabilities were calculated on the basis of the pre-test probabilities of this patient group. All patients had cytologically or histologically proven cancer. The prevalence of additional lymph node metastases on PET in staging examinations was 19.6% (11/56), and that of locoregional recurrence on restaging PET was 28.6% (12/42). In the primary assessment of patients, PET had positive and negative likelihood ratios of 3.9 (2.56-5.93) and 0.24 (0.14-0.41), respectively. Disease probabilities were therefore 49.4% for a positive test result and 5.7% for a negative test result. In the assessment of recurrence these values were 3.96 (2.8-5.6) and 0.16 (0.1-0.25), resulting in probabilities of 49.7% and 3.8%. PET evaluation for involvement of lymph nodes had positive and negative likelihood ratios of 17.26 (10.9-27.3) and 0.19 (0.13-0.27) for primary assessment and 11.0 (2.93-41.24) and 0.14 (0.01-1.88) for detection of recurrence. The probabilities were 81.2% and 4.5% for primary assessment and 73.3% and 3.4% for assessment of recurrence. It is concluded that in this clinical setting the main advantage of PET is the ability to reliably rule out the presence of disease in both staging and restaging. Further research is required to derive probabilities for individual patients from sequential testing as applied in the diagnostic work-up of patients with head and neck cancer.