BACKGROUND: Microsatellite-stable metastatic colorectal cancer is typically unresponsive to immunotherapy. This phase 3 study was designed to assess atezolizumab plus cobimetinib in metastatic colorectal cancer. Here, we report the comparison of atezolizumab plus cobimetinib or atezolizumab monotherapy versus regorafenib in the third-line setting. METHODS: IMblaze 370 is a multicentre, open-label, phase 3, randomised, controlled trial, done at 73 academic medical centres and community oncology practices in 11 countries. Patients aged at least 18 years with unresectable locally advanced or metastatic colorectal cancer, baseline Eastern Cooperative Oncology Group performance status of 0-1, and disease progression on or intolerance to at least two previous systemic chemotherapy regimens were enrolled. We used permuted-block randomisation (block size four) to assign patients (2:1:1) via an interactive voice and web response system to atezolizumab (840 mg intravenously every 2 weeks) plus cobimetinib (60 mg orally once daily for days 1-21 of a 28-day cycle), atezolizumab monotherapy (1200 mg intravenously every 3 weeks), or regorafenib (160 mg orally once daily for days 1-21 of a 28-day cycle). Stratification factors were extended RAS status (wild-type vs mutant) and time since diagnosis of first metastasis (<18 months vs ≥18 months). Recruitment of patients with high microsatellite instability was capped at 5%. The primary endpoint was overall survival in the intention-to-treat population. Safety was assessed in the population of patients who received at least one dose of their assigned treatment. IMblaze370 is ongoing and is registered with ClinicalTrials.gov, number NCT02788279. FINDINGS: Between July 27, 2016, and Jan 19, 2017, 363 patients were enrolled (183 patients in the atezolizumab plus cobimetinib group, 90 in the atezolizumab group, and 90 in the regorafenib group). At data cutoff (March 9, 2018), median follow-up was 7·3 months (IQR 3·7-13·6). Median overall survival was 8·87 months (95% CI 7·00-10·61) with atezolizumab plus cobimetinib, 7·10 months (6·05-10·05) with atezolizumab, and 8·51 months (6·41-10·71) with regorafenib; the hazard ratio was 1·00 (95% CI 0·73-1·38; p=0·99) for the combination versus regorafenib and 1·19 (0·83-1·71; p=0·34) for atezolizumab versus regorafenib. Grade 3-4 adverse events were reported in 109 (61%) of 179 patients in the atezolizumab plus cobimetinib group, 28 (31%) of 90 in the atezolizumab group, and 46 (58%) of 80 in the regorafenib group. The most common all-cause grade 3-4 adverse events in the combination group were diarrhoea (20 [11%] of 179), anaemia (ten [6%]), increased blood creatine phosphokinase (12 [7%]), and fatigue (eight [4%]). Serious adverse events were reported in 71 (40%) of 179 patients in the combination group, 15 (17%) of 90 in the atezolizumab group, and 18 (23%) of 80 in the regorafenib group. Two treatment-related deaths occurred in the combination group (sepsis) and one in the regorafenib group (intestinal perforation). INTERPRETATION: IMblaze370 did not meet its primary endpoint of improved overall survival with atezolizumab plus cobimetinib or atezolizumab versus regorafenib. The safety of atezolizumab plus cobimetinib was consistent with those of the individual drugs. These results underscore the challenge of expanding the benefit of immunotherapy to patients whose tumours have lower baseline levels of immune inflammation, such as those with microsatellite-stable metastatic colorectal cancer. FUNDING: F Hoffmann-La Roche Ltd/Genentech Inc.