To develop a clinical model for diagnosis of bacterial urinary tract infection, we conducted a prospective study on 266 dysuric young women, 147 of whom had urinary tract infections. Five variables were found to be significant and independent correlates to bacterial urinary tract infection on logistic regression analysis: sexual activity, absence of vaginal discharge, short duration of complaints, leukocyturia, and hematuria. An algorithm combining the logistic model and a Gram-stained urine specimen, which was used in only a third of our patients, afforded a sensitivity of 86% and a specificity of 84%. The algorithm was validated in a second set of 166 dysuric women, 77 of whom had urinary tract infections. The algorithm led to a diagnosis of bacterial urinary tract infection with a sensitivity of 91% and specificity of 79%; the only laboratory test needed except for urinalysis was a Gram's stain of urine, obtained for 30% of the patients.