Abstract Background Physical inactivity is an established risk factor for several cancers of the digestive system and female reproductive organs, but the evidence for liver cancers is less conclusive. Methods The aim of this study was to synthesize prospective observational studies on the association of physical activity and liver cancer risk by means of a systematic review and meta-analysis. We searched Medline, Embase, and Scopus from inception to January 2019 for prospective studies investigating the association of physical activity and liver cancer risk. We calculated mean hazard ratios (HRs) and 95% confidence intervals (CIs) using a random-effects model. We quantified the extent to which an unmeasured confounder or an unaccounted selection variable could shift the mean hazard ratio to the null. Results Fourteen prospective studies, including 6,440 liver cancers, were included in the systematic review and meta-analysis. The mean hazard ratio for high compared with low physical activity was 0.75 (95% CI = 0.63 to 0.89; 95% prediction interval = 0.52 to 1.07; I² = 64.2%). We estimated that 67.6% (95% CI = 56.6% to 78.5%) of all true effect estimates would have a hazard ratio less than 0.8. Bias analysis suggested than an unobserved confounder would have to be associated with a 1.99-fold increase in the risk of physical activity or liver cancer to explain away the observed mean hazard ratio. An unaccounted for selection variable would have to be related to exposure and endpoint with a relative risk of 1.58 to explain away the mean hazard ratio. Conclusions Physical activity is inversely related to the risk of liver cancer. Further studies with objectively measured physical activity and quasi-experimental designs addressing confounding are needed.