The development of human cancer is a multistep process characterized by the accumulation of genetic and epigenetic alterations that drive or reflect tumour progression. These changes distinguish cancer cells from their normal counterparts, allowing tumours to be recognized as foreign by the immune system. However, tumours are rarely rejected spontaneously, reflecting their ability to maintain an immunosuppressive microenvironment. Programmed death-ligand 1 (PD-L1; also called B7-H1 or CD274), which is expressed on many cancer and immune cells, plays an important part in blocking the 'cancer immunity cycle' by binding programmed death-1 (PD-1) and B7.1 (CD80), both of which are negative regulators of T-lymphocyte activation. Binding of PD-L1 to its receptors suppresses T-cell migration, proliferation and secretion of cytotoxic mediators, and restricts tumour cell killing. The PD-L1-PD-1 axis protects the host from overactive T-effector cells not only in cancer but also during microbial infections. Blocking PD-L1 should therefore enhance anticancer immunity, but little is known about predictive factors of efficacy. This study was designed to evaluate the safety, activity and biomarkers of PD-L1 inhibition using the engineered humanized antibody MPDL3280A. Here we show that across multiple cancer types, responses (as evaluated by Response Evaluation Criteria in Solid Tumours, version 1.1) were observed in patients with tumours expressing high levels of PD-L1, especially when PD-L1 was expressed by tumour-infiltrating immune cells. Furthermore, responses were associated with T-helper type 1 (TH1) gene expression, CTLA4 expression and the absence of fractalkine (CX3CL1) in baseline tumour specimens. Together, these data suggest that MPDL3280A is most effective in patients in which pre-existing immunity is suppressed by PD-L1, and is re-invigorated on antibody treatment.