Equivocal findings on crime as a deterrent for physical activity may be due to effects of geographic scale on exposure measurement. To investigate this hypothesis, physical activity was measured in 203,883 Australians and linked to standardised crime counts within small (‘Census Collection Districts’; approx. 330 residents) and larger areas (‘Statistical Local Areas’; approx. 32,000 residents). A median rate ratio of 2.26 indicated substantive geographic variation in moderate-to-vigorous physical activity (MVPA). Adjusting for confounders, multilevel negative binomial regression reported lower MVPA with more crime consistently in small, but not in larger areas. Reducing small pockets of local crime may encourage more physically active lifestyles.