OBJECTIVE--To establish the incidence of maxillary sinusitis in general practice and the predictive value of symptoms and signs. DESIGN--Population based study. SETTING--9 general practices with 15,220 patients aged 15 years and older on the list. PATIENTS--400 patients with 441 episodes in whom practitioners intended to confirm or to exclude sinusitis. MAIN OUTCOME MEASURES--Results of ultrasonography and signs and symptoms associated with positive results. RESULTS--212 of the 441 episodes were confirmed by ultrasonography. 15.7 episodes occurred per 1000 adults per year. The five symptoms beginning with common cold (beta coefficient = 1.035), purulent rhinorrhoea (0.996), pain at bending (0.950), unilateral maxillary pain (0.640), and pain in teeth (0.606) were associated with positive results on ultrasonography. General practitioners' clinical diagnoses were correct in 177 episodes, false positive in 88, false negative in 22, and uncertain in 154. With an algorithm using the five weighted symptoms 243 of the diagnoses would have been correct, but 110 would remain uncertain and 44 cases would have been missed. CONCLUSION--The five symptoms algorithm would improve diagnostic accuracy of general practitioners, but incorrect and uncertain diagnoses cannot be avoided.