OBJECTIVE: To investigate the association of radiographic progression defined by Prostate Cancer Working Group (PCWG)-2 guidelines and overall survival (OS) in men with metastatic castration-resistant prostate cancer (mCRPC). PATIENTS AND METHODS: Two trials that used PCWG-2 guidelines to define progression were analysed: a randomized phase II trial (n = 221) comparing first-line docetaxel-prednisone plus AT-101 or placebo, and a phase III trial (n = 873) comparing prednisone plus sunitinib or placebo after docetaxel-based chemotherapy. Cox proportional hazards regression models were used to estimate the association of radiographic progression with OS. Landmark analyses compared progressing patients with those who had not progressed. Sub-analyses compared patients removed from trial for progression vs other reasons. RESULTS: An increased risk of death was seen for radiographic progression at landmark times from 6 to 12 months with docetaxel-based therapy (hazard ratio [HR] >1.7 at all time-points). An increased risk of death was also seen with post-docetaxel prednisone alone or with sunitinib for progression at landmark times from 2 to 8 months (HR >2.7 at all time-points). Kendall's τ was 0.50 (P < 0.001) in the setting of docetaxel-based therapy and 0.34 (P < 0.001) in the post-docetaxel setting for association between radiographic progression and death amongst patients with both events. Removal from study due to radiographic progression was associated with a significantly lower OS compared with removal for other reasons in both trials. Limitations of a retrospective analysis apply and there was no central radiology review. CONCLUSIONS: Radiographic progression by PCWG-2 criteria was significantly associated with OS in patients with mCRPC receiving first-line docetaxel-based chemotherapy or post-docetaxel therapy. With external validation as a surrogate endpoint in trials showing survival benefits, the use of radiographic progression-free survival may expedite drug development in mCRPC, which has been hampered by the lack of intermediate endpoints.