Abstract Purpose: This randomized, multicenter, open-label, phase Ib/II study assessed durvalumab and tremelimumab in combination or as monotherapy for chemotherapy-refractory gastric cancer or gastroesophageal junction (GEJ) cancer. Patients and Methods: Second-line patients were randomized 2:2:1 to receive durvalumab plus tremelimumab (arm A), or durvalumab (arm B) or tremelimumab monotherapy (arm C), and third-line patients received durvalumab plus tremelimumab (arm D). A tumor-based IFNγ gene signature was prospectively evaluated as a potential predictive biomarker in second- and third-line patients receiving the combination (arm E). The coprimary endpoints were objective response rate and progression-free survival (PFS) rate at 6 months. Results: A total of 113 patients were treated: 6 in phase Ib and 107 (arm A, 27; arm B, 24; arm C, 12; arm D, 25; arm E, 19) in phase II. Overall response rates were 7.4%, 0%, 8.3%, 4.0%, and 15.8% in the five arms, respectively. PFS rates at 6 months were 6.1%, 0%, 20%, 15%, and 0%, and 12-month overall survival rates were 37.0%, 4.6%, 22.9%, 38.8%, and NA, respectively. Treatment-related grade 3/4 adverse events were reported in 17%, 4%, 42%, 16%, and 11% of patients, respectively. Conclusions: Response rates were low regardless of monotherapy or combination strategies. No new safety signals were identified. Including use of a tumor-based IFNγ signature and change in baseline and on-treatment circulating tumor DNA are clinically feasible and may be novel strategies to improve treatment response in this difficult-to-treat population.