ObjectiveThere are some indications of regional differences in the association between fish consumption and clinical outcomes. We aimed to test the linear and potential non-linear dose–response relationships between fish consumption and risk of all-cause and cardiovascular (CVD) mortality, and possible confounding by region.DesignSystematic review and dose–response meta-analysis.SettingSystematic search using PubMed and Scopus, from inception up to September 2016.SubjectsProspective observational studies reporting the estimates of all-cause and CVD mortality in relation to three or more categories of fish intake were included. Random-effects dose–response meta-analysis was conducted.ResultsFourteen prospective cohort studies (ten publications) with 911 348 participants and 75 451 incident deaths were included. A 20 g/d increment in fish consumption was significantly and inversely associated with the risk of CVD mortality (relative risk=0·96; 95 % CI 0·94, 0·98; I2=0 %, n 8) and marginally and inversely associated with the risk of all-cause mortality (relative risk=0·98; 95 % CI 0·97, 1·00; I2=81·9 %, n 14). Subgroup analysis resulted in a significant association only in the subgroup of Asian studies, compared with Western studies, in both analyses. Analysis of Western studies suggested a nearly U-shaped association, with a nadir at fish consumption of ~20 g/d in analysis of both outcomes. Meanwhile, the associations appeared to be linear in Asian studies.ConclusionsThere was potential evidence of regional differences in the association between fish consumption and mortality. It may be helpful to examine the associations by considering types of fish consumed and methods of fish preparation.