Issues AddressedEvidence on how COVID‐19 lockdowns impacted physical activity (PA) is mixed. This study explores changes in PA following initial mobility restrictions, and their subsequent relaxation, in a sample of Sydney (Australia) residents using a natural experiment methodology.MethodsParticipants' health and travel behaviours were collected pre‐pandemic in late 2019 (n = 1937), with follow‐up waves during the pandemic in 2020 (n = 1706) and 2021 (n = 1514). Linear mixed‐effects models were used to analyse changes in weekly duration of PA between the three waves.ResultsCompared with pre‐pandemic, average weekly PA increased in 2021 by 42.6 min total PA (p = .001), 16 min walking PA (p = .02), and 26.4 min moderate–vigorous PA (MVPA) (p = .003). However, average weekly sessions of PA decreased in 2020 and remained lower in 2021. For participants who were sufficiently active in 2019, weekly total PA (−66.3 min) MVPA (−43.8 min) decreased in 2020 compared to pre‐pandemic. Conversely, among participants who were insufficiently active in 2019, average weekly PA increased in both 2020 (total PA, +99.1 min; walking PA, +46.4 min; MVPA +52.8 min) and 2021 (total PA, +117.8 min; walking PA, +58.4 min; MVPA +59.2 min), compared to 2019. Participants who did more work from home increased their average weekly total PA in 2021 compared to pre‐pandemic (+45.3 min).ConclusionThese findings reveal the complex variability in PA behaviour brought about by the pandemic.So What?Strategies to support the population in achieving sufficient PA must focus on maintaining an appetite for PA as we move out of the pandemic and on promoting more frequent PA sessions.