Background: Immune checkpoint inhibitors (ICIs) have become the preferred treatment for advanced hepatocellular carcinoma (HCC). Surrogate endpoints including overall response rate (ORR) and progression-free survival (PFS) are preferred outcome measures in early-phase trials. However, ICIs can induce atypical patterns of response and progression, calling into question the validity of surrogate endpoints as predictive of overall survival (OS) benefit. We performed a post-hoc analysis of a large, randomized phase 3 clinical trial (IMbrave150) and cross-sectional analysis of a multi-institution real-world environment (RWE) to assess the OS surrogacy of ORR and PFS using Response Evaluation Criteria in Solid Tumors (RECIST), modified RECIST (mRECIST), or immune-modified RECIST (imRECIST) in patients treated with ICIs. Methods: The IMbrave150 cohort in our analysis included 279 out of 336 patients treated with atezolizumab and bevacizumab in the intent-to-treat population. The RWE cohort included 328 patients with Child-Pugh A or B disease treated in the first or subsequent-line setting with anti-PD-1/L1 therapies. RECIST, mRECIST, and imRECIST ORR and PFS data for the IMbrave150 cohort was determined by investigators and central review, while RECIST ORR and PFS data for the RWE cohort was determined by treating providers or retrospective review. Results: In the IMbrave150 and RWE cohorts, Child-Pugh A patients treated in the first-line setting with CR/PR showed greater OS versus patients with SD/PD (IMbrave150: HR 0.16, 95% CI 0.10-0.26; RWE: HR 0.25, 95% CI 0.15-0.43). CR was associated with the greatest OS benefit, followed by PR, SD, and PD. Patients treated in the second-line setting or with Child-Pugh B disease showed that ORR was able to discriminate OS benefit but hazard ratios were closer to the null. In both cohorts, there was a positive rank correlation between RECIST PFS and OS (IMbrave150: Kendall's τ = 0.44, p < 0.001); RWE: Kendall's τ = 0.52, p < 0.001). However, only 31-46% of the variance in OS rankings was explained by PFS rankings. mRECIST PFS and imRECIST PFS showed a positive rank correlation with OS, with an explained variance of 30% and 38% for mRECIST and imRECIST, respectively. Conclusions: Our results suggest that RECIST ORR is a robust surrogate endpoint in HCC treated with ICIs. In contrast, the explained variance between RECIST PFS and OS in both the IMbrave and RWE cohorts was limited. This may reflect the impact of ICIs on response to subsequent therapies which may extend OS benefit. Although prior studies have proposed that modifications to RECIST may lead to more accurate surrogate markers, mRECIST and imRECIST ORR or PFS in HCC treated with ICIs were not associated with greater predictive performance compared to RECIST.