Objective:To define the factors associated with diagnosis of toxoplasmic encephalitis (TE) in AIDS patients; and to establish a rational procedure for the clinician faced with a decision concerning empiric antitoxoplasma therapy.Design:A 15-month prospective multicentre cohort study in France.Methods:One hundred and eighty-six consecutive HIV-positive inpatients undergoing empiric antitoxoplasma therapy for a first episode of presumed TE were monitored. The clinician's initial estimation of the probability of response to antitoxoplasma therapy was recorded. In addition, a validation committee classified cases as TE or non-TE.Results:Among the 186 patients, the following variables were significantly more frequent in TE (n = 113) than non-TE (n = 73) patients: fever (59% versus 40%), headache (55% versus 33%), seizures (22% versus 11%), suggestive lesions on the brain scan (98% versus 76%), positive Toxoplasma serology (97% versus 71%). Median CD4+ lymphocyte count was significantly higher in TE than in non-TE (27 x 106/l versus 11 x 106/l). The rate of TE in patients on systemic antiprotozoal prophylaxis at entry was 43% as compared with 75% in patients without previous prophylaxis. Pre-therapy estimation of response to empiric therapy was highly correlated with final diagnosis. Multivariate logistic regression analysis showed that the following variables contributed independently to the diagnosis of TE: clinician's estimation of response to treatment at entry > 75%; absence of systemic antiprotozoal prophylaxis; seizures; headache; suggestive lesions on CT or MRI brain scan; and positive Toxoplasmaserology.Conclusions:A linear logistic model is proposed which uses significant variables, which are readily available. This model gives good accuracy to classify suspected cases of TE.