Objective: To examine the association between exercise during pregnancy and preterm birth (PTB). Design: Retrospective cohort analysis. Setting: Jinan, China. Population: Singleton live births from December 2018 to December 2019. Methods: Questionnaires contained items about physical exercise (frequency, time, and primary exercise patterns) during pregnancy and each trimester. Adjusted odds ratios (OR) were estimated using logistic regression. Variable selection for the multivariate models was guided by the directed acyclic graph. The median effect was analyzed by the sequential test. Main Outcome Measures: PTB. Results: The prevalence of PTB in this study was 4.38% (285/6501). The adjusted OR (95% CI) for the risk of PTB related to exercise during pregnancy was 0.74 (0.58-0.95). During the 1st and 2nd trimesters, the ORs (95% CI) for 2.5 to 7 hours of exercise per week were 0.77 (0.59-0.99) and 0.74 (0.57-0.96). During the 3rd trimester, the ORs (95% CI) for 2.5 to 7 hours and more than 7 hours of exercise per week were 0.74 (0.56-0.96) and 0.65 (0.44-0.94). After stratifying the subjects, the association was only found among subjects without pregnancy complications. Pregnancy complications partially mediated (52.40%) the relationship between exercise during pregnancy and PTB. Conclusions: Exercise during pregnancy was a protective factor of PTB for women without pregnancy complications. 2.5 to 7 hours of exercise (like walking) per week may be appropriate in three trimesters of pregnancy, and the time could be extended in the 3rd trimester. Keywords: Preterm birth; Exercise; Pregnancy; Trimester; Mediation analysis