Background: A sensitive liquid chromatography tandem-mass spectrometry (LC-MS/MS) method was used to monitor serum testosterone levels in castrated prostate cancer patients. We subsequently performed an observational and retrospective study to estimate the within- and between-subject biological variation of these patients. Methods: In total, 474 samples from 72 prostate cancer patients in the Netherlands receiving either chemical castration (CAS) or castration plus enzalutamide (ENZA) treatment were selected for data analysis. ANOVA was performed to estimate analytical variation (CV A) and within-patient variation (CV I). A nested ANOVA was applied to estimate between-patient variation (CV G). From these data, the reference change value (RCV) and analytical performance specifications (APS) were calculated. Results: Testosterone levels were significantly higher in the ENZA group (0.318 vs. 0.191 nmol/L, p < 0.005) than the CAS group. Overall, variation components were estimated at 6.1%, 24.6% and 60.3% for CV A, CV I and CV G, respectively. Both groups showed high individuality (<0.6). The RCV was 70.3% for all patients. Desirable APS were 12.3% for imprecision, 16.3% for bias and 26.4% for total error. Conclusion: The generated APS are valuable for sensitive testosterone assays and the high individuality indicates that castrated testosterone levels can be studied as a predictive or prognostic biomarker in prostate cancer patients.