BACKGROUND Sore throat is very common in general practice and is usually caused by viral infection. Nevertheless, up to 95% of patients may be treated with antibiotics. Previous diagnostic systems have not transferred well from one area to another because of an inability to allow for changing prevalence of streptococcus. OBJECTIVE To measure the occurrence rates of symptoms and signs in sore throat patients with and without streptococcal infection, and to develop a Bayesian scoring system which is easily adapted for prevalence to predict if patients have bacterial infection. METHODS Occurrence rates of symptoms and signs were measured for 206 patients with sore throat symptoms over a 3-year period. Bayesian probability scores (B-scores) for each data item were calculated from the ocurrence rates in the patients with positive throat cultures for group A streptococci and the rates in patients with negative throat cultures. The B-score values were then used to predict the probability of positive culture for each patient. RESULTS The streptococcal throat B-score system predicted positive culture with a sensitivity of 71% and a specificity of 71%. In comparison, the unaided general practitioners predicted infection with a sensitivity of 61% and a specificity of 65%. If the B-score prediction had been used to decide on treatment, more patients with streptococci present on culture would have been treated with antibiotic (71% instead of 68%) and appreciably fewer patients with negative streptococcal cultures would have been treated (29% instead of 59%). CONCLUSIONS Use of the B-score system could result in significant savings in unnecessary antibiotic prescription, and unnecessary throat swab cultures, while achieving better levels of treatment.