The relationship between physical activity (PA) before cancer diagnosis and all cancer mortality among the general population is not well defined because of inconsistent results from published studies. Thus, the lack of a meta-analysis that addresses that issue prompted the current report. We conducted a literature search of PubMed and Web of Science to identify all relevant epidemiological studies published before February 28, 2015. We performed categorical and dose-response meta-analyses to evaluate and quantify the association between pre-diagnosis PA and all cancer mortality. A total of 32 prospective cohort studies involving 59,362 cancer deaths were included in this meta-analysis. The pooled relative risks (RRs) of all cancer mortality were 0.80 [95% confidence interval (CI) = 0.76-0.85)] for highest versus lowest PA group and 0.85 (95% CI = 0.82-0.88) for PA versus non/occasional PA group. Dose-response analysis showed that the increment in pre-diagnosis PA level was associated with a decreased risk of cancer death continuously. Moreover, an increment of 10 MET-h/week was related to a 7% lower risk for all cancer mortality (RR = 0.93, 95% CI = 0.91-0.95). In conclusion, the present meta-analysis provides evidence of an inverse association between pre-diagnosis PA and all cancer mortality among the general population. High-quality epidemiological studies that employ standardized PA assessments and unified definitions of PA levels should be developed in future.