OBJECTIVE: To analyse individual variations in serum testosterone level, the cumulative rate of 'breakthrough' increases over castrate levels, and to evaluate whether the increases can be predicted. PATIENTS AND METHODS: Serum testosterone levels were determined every 6 months over 3 years in 73 consecutive patients with prostate cancer who were medically castrated, prospectively enrolled in a single tertiary academic centre. Patients recruited for this study were being treated with a 3-monthly depot of luteinizing hormone-releasing hormone agonist over 6-48 months. Serum testosterone was measured using a chemiluminescent assay with a lower sensitivity level of 15 ng/dL and interassay coefficient of variation of 25% at low testosterone concentrations. RESULTS: Individual variations could not be explained by the interassay variation coefficient in 26% of the patients. The rate of breakthrough increases >50 ng/dL increased from 12.3% at the first determination to 24.7% at the third, then remaining stable. The rate of breakthrough increases of 20-50 ng/dL increased from 27.4% at the first determination to 31.5% at the second, and then remained stable. A first determination of <20 ng/dL provided an 11.4% probability for future increases of >50 ng/dL, with a 5.7% probability if two consecutive determinations were <20 ng/dL and a null probability when three consecutive determinations were <20 ng/dL. CONCLUSIONS: Individual variations in serum testosterone level cannot be explained by the coefficient of variation of the assay in a quarter of patients who are medically castrated. Breakthrough increases over castrate levels increase over time and those of >50 ng/dL can be predicted from the previous levels.