PURPOSE Neoadjuvant gemcitabine and cisplatin (GC) followed by radical cystectomy (RC) is standard for patients with muscle-invasive bladder cancer (MIBC). On the basis of the activity of atezolizumab (A) in metastatic BC, we tested neoadjuvant GC plus A for MIBC. METHODS Eligible patients with MIBC (cT2-T4aN0M0) received a dose of A, followed 2 weeks later by GC plus A every 21 days for four cycles followed 3 weeks later by a dose of A before RC. The primary end point was non–muscle-invasive downstaging to < pT2N0. RESULTS Of 44 enrolled patients, 39 were evaluable. The primary end point was met, with 27 of 39 patients (69%) < pT2N0, including 16 (41%) pT0N0. No patient with < pT2N0 relapsed and four (11%) with ≥ pT2N0 relapsed with a median follow-up of 16.5 months (range: 7.0-33.7 months). One patient refused RC and two developed metastatic disease before RC; all were considered nonresponders. The most common grade 3-4 adverse event (AE) was neutropenia (n = 16; 36%). Grade 3 immune-related AEs occurred in five (11%) patients with two (5%) requiring systemic steroids. The median time from last dose of chemotherapy to surgery was 7.8 weeks (range: 5.1-17 weeks), and no patient failed to undergo RC because of AEs. Four of 39 (10%) patients had programmed death-ligand 1 (PD-L1)–positive tumors and were all < pT2N0. Of the patients with PD-L1 low or negative tumors, 23 of 34 (68%) achieved < pT2N0 and 11 of 34 (32%) were ≥ pT2N0 ( P = .3 for association between PD-L1 and < pT2N0). CONCLUSION Neoadjuvant GC plus A is a promising regimen for MIBC and warrants further study. Patients with < pT2N0 experienced improved relapse-free survival. The PD-L1 positivity rate was low compared with published data, which limits conclusions regarding PD-L1 as a predictive biomarker.