Individuals frequently fail to follow through on their often admirable health behavior intentions, with adverse consequences for individual and population health. More effective behavior change interventions are required to address this issue. We take the first steps towards achieving this aim by using dual-self models to gain crucial insight into designing such interventions. Present bias can explain individuals not following through on their intentions, but the quasi-hyperbolic model gives no insight into the underlying determinants of such present bias. We show theoretically that the dual-self models suggest that present bias may be determined by willpower resources, cognitive load and affective intensity. We do this by first adapting the quasi-hyperbolic and dual-self models to a health behavior context and then comparing the adapted models. We also generate predictions to empirically test the models against each other. We outline the design of a field study to test these predictions and our findings in relation to present bias.