Testosterone levels were measured in blood and urine of 35 premenopausal metastatic breast cancer patients before starting therapy with the gonadotrophin-releasing hormone (GnRH) analogue, goserelin. The aim of the study was to verify the reliability of testosterone measurement as a marker of prognosis. The time interval between starting therapy and progressive disease (time to progression) was chosen to assess prognosis. Univariate and multivariate analysis showed that only urinary testosterone levels were significantly associated with time to progression (Wald test 6.66, P = 0.01 for univariate and Wald test 7.93, P = 0.0049 for multivariate analysis), whereas no association was found for testosterone in blood. A statistical model is proposed to evaluate probability of progressive disease in relation to testosterone values in urine at different times. According to the model, the probability of progression decreases with increasing urinary testosterone values.