Background: Surrogate endpoints for prostate cancer-specific mortality after curative primary treatment are not well established. We sought to assess time to biochemical failure (TTBF) and prostate-specific antigen doubling time (PSADT) after failure of curative treatment as candidates for this endpoint. Methods: PSA and survival data from the Trans-Tasman Radiation Oncology Group (TROG) 96.01 trial were used to assess surrogate candidates. Between June 28, 1996, and Feb 16, 2000, 802 eligible men with locally advanced prostate cancer were randomly allocated to prostatic irradiation alone, or to 3 or 6 months of maximum short-term androgen deprivation (STAD) before and during radiation. Successful surrogates were required to satisfy the Prentice criteria and to predict the trial finding. The TROG 96.01 trial is registered with the Australian New Zealand Clinical Trials Registry, number ACTRN12607000237482. Findings: 6 months of STAD was shown to significantly decrease prostate cancer-specific mortality compared with radiation alone, but 3 months of STAD did not result in a decrease. Relative to radiation alone, the hazard ratio of prostate cancer-specific mortality from randomisation was 0·95 (95% CI 0·63–1·41; p=0·79) in the 3-month STAD treatment arm and 0·56 (0·36–0·88; p=0·01) in the 6-month arm. PSADT predicted the trial finding and satisfied all four Prentice criteria at the cutpoints of less than 12 months and less than 15 months, with proportion of treatment effect ratios between 0·36 and 0·56. Time to biochemical failure was better than PSADT at predicting the trial finding and satisfying all four Prentice criteria at cutpoints of less than 1·5, less than 2, and less than 2·5 years, with proportion of treatment effect ratios between 0·45 and 0·64. Interpretation: This study provides proof of principle that TTBF and PSADT can be useful as surrogate endpoints for prostate cancer-specific mortality and offer potential to substantially reduce follow up in clinical trials. These endpoints now require assessment in multi-trial meta-analyses before use in clinical trials.