Purpose: The impact of physical activity (PA) on Quality of Well-being (QWB) scores, and the effects of PA on Markov models considering specific diseases or all-cause mortality were studied. Methods: We combined national PA and quality of life (QOL) data to derive PA-related QWB scores. Well subjects (defined as free of the five inactivity-related diseases modeled) were stratified across four PA levels according to public health recommendations. Three multiple regressions were fit. First, using QWB data from persons with disease, QOL was fit as a function of age, gender and type of disease. Next, using data from those who were well, QOL was fit as a function of age, gender and PA level. Finally, using data for all persons, QOL was again fit as a function of age, gender and exercise level. The first two regressions were used to estimate QOL in our disease-specific model, which simulated the effect of PA on disease, and the subsequent effect of disease on QOL, mortality and medical costs. The third regression was used in an all-cause mortality model which simulated the direct effect of PA on QWB, mortality and medical costs. Both models were then used to estimate the gain in QALYs from select interventions that improved PA, and compared them. Results: Among the well, there was a strong relationship between QOL and PA level. For example, 40-year old women who were highly active had a mean QOL of 0.818,_while those who met PA recommendations, were irregularly active, or inactive, reported mean QOLs of 0.807, 0.796, and 0.791 respectively. In the disease-specific model, the average gain in QALYs from an evaluated community-wide campaign strategy, was 14.816, while in the all-cause mortality model, the same intervention had an average gain in QALYs of 15.550. Discussion: Well adults were found to have PA-associated differences in health and QOL. Models that do not take into account the positive effect of healthy behavior among the well may underestimate the effect of behavior change on QALYs. When QOL, mortality and cost data are available by behavior level, designing “all-cause” models that take into account the direct effect of healthful behaviors on outcomes, without simulating the intervening effect of disease, may yield similar and equally accurate results.