Previous studies link lower physical activity with incident Parkinson's disease (PD) but rely on self-reported data and fail to address reverse causation. This study used accelerometer-derived daily step count, an objective measure of physical activity, to examine its association with incident PD in the UK Biobank, within successive periods of follow-up. PD cases were identified through hospital inpatient and death data, and associated with machine learning-derived step counts using Cox regression models, adjusted for age, sex, demographic and lifestyle factors. For 94,696 valid participants and a median follow-up of 7.9 years, 407 incident PD cases occurred. Every 1000 steps higher were associated with 8% lower risk of PD (hazard ratio 0.92, 95% confidence interval 0.89-0.94). However, this association was no longer statistically significant when excluding follow-up periods closer to the time of accelerometer wear, suggesting that low activity may be an early marker, but not a risk factor for PD.