To assess the association of fish consumption with risk of dementia and its dose-response relationship, and investigate variations in the association among low-, middle- and high-income countries. A new community-based cross-sectional study and a systematic literature review.SettingsUrban and rural communities in China; population-based studies systematically searched from worldwide literature. Chinese adults aged ≥60 years in six provinces (n 6981) took part in a household health survey of dementia prevalence and risk factors. In addition, 33 964 participants from eleven published and eligible studies were included in the systematic review and meta-analysis. In the new study in China, 326 participants were diagnosed with dementia (4·7 %); those who consumed any amount of fish in the past two years v. those who consumed no fish had reduced risk of dementia (adjusted OR=0·73, 95 % CI 0·64, 0·99), but the dose-response relationship was not statistically significant. The meta-analysis of available data from the literature and the new study showed relative risk (RR) of dementia of 0·80 (95 % CI 0·74, 0·87) for people with fish consumption; the impact was similar among countries with different levels of income. Pooled dose-response data revealed RR (95 % CI) of 0·84 (0·72, 0·98), 0·78 (0·68, 0·90) and 0·77 (0·61, 0·98) in people with low, middle and high consumption of fish, respectively. Corresponding figures for Alzheimer's disease were 0·88 (0·74, 1·04), 0·79 (0·65, 0·96) and 0·67 (0·58, 0·78), respectively. Greater consumption of fish is associated with a lower risk of dementia. Increasing fish consumption may help prevent dementia worldwide regardless of income level.