Background: Physical activity (PA) and discretionary screen time (DST; television and computer use during leisure) are both associated with obesity risk, but little longitudinal evidence exists on their combined influence. This study examined the independent and joint associations of changes in PA and DST with incident obesity, body mass index (BMI) and waist circumference (WC). Methods: We analysed the data of individuals aged 40–69 years from the UK Biobank, a large-scale, population-based prospective cohort study. PA was measured using the International Physical Activity Questionnaire and DST was defined as the total of daily TV viewing and non-occupational computer use. Changes in PA and DST over time were defined using departure from sex-specific baseline tertiles and categorised as worsened (PA decreased/DST increased), maintained, and improved (PA increased/DST decreased). We then used each exposure change to define a joint PA-DST change variable with nine mutually exclusive groups. We used multivariable adjusted mixed-effects linear and Poisson models to examine the independent and joint associations between PA and DST changes with BMI and WC and incident obesity, respectively. Development of a BMI ≥ 30 kg/m2 was defined as incident obesity. Results: Among 30,735 participants, 1,628 (5.3%) developed incident obesity over a mean follow-up of 6.9 (2.2) years. In the independent association analyses, improving PA (Incident Rate Ratio (IRR) 0.46 (0.38–0.56)) was associated with a lower risk of incident obesity than maintaining PA, maintaining DST, or improving DST. Compared to the referent group (both PA and DST worsened), all other combinations of PA and DST changes were associated with lower incident obesity risk in the joint association analyses. We observed substantial beneficial associations in the improved PA groups, regardless of DST change [e.g., DST worsened (IRR 0.31 (0.21–0.44)), maintained (IRR 0.34 (0.25–0.46)), or improved (IRR 0.35 (0.22–0.56)]. The most pronounced decline in BMI and WC was observed when PA was maintained or improved and DST was maintained. Conclusion: We found that improved PA had the most pronounced beneficial associations with incident obesity, irrespective of DST changes. Improvements in PA or DST mutually attenuated the deleterious effects of the other behaviour’s deterioration.