Demographic, clinical, and laboratory data from 200 consecutive patients with acute brucellosis were analysed with univariate and multivariate methods to identify correlates of relapse. A risk score for predicting relapse was then calculated by using Cox proportional hazard model. The independent predictors of relapse were temperature of 38.3 degrees C or higher, positive blood cultures at baseline, and the duration of symptoms before treatment <10 days. Stratification according to the risk score demonstrated that rates of relapse were significantly different between risk groups (P<0.0001). The low-risk group had a 4.5% probability (6 of 135) of relapse at 12 months. In contrast, relapse was present in 15 of 47 patients in the medium-risk group (P<0.0017); and in 12 of 18 patients in the high-risk group (P<0.0001). This study provides a rational basis for estimating the risk of relapse in patients with acute brucellosis, and may be helpful in deciding what subjects might benefit from extra attention.