Sequential sampling models, such as Drift Diffusion Models (DDM), are used to understand choice processes. In recent years, these models have become increasingly elaborate, for example by considering the choice options’ underlying attributes, or attentional dynamics during decision making. Here we compare different DDMs of varying complexity to elucidate the mechanisms underlying hunger-driven effects in dietary choice. The models were fit to data of 70 participants completing a binary food choice task in hungry and sated states (within-subject design) while their eye-movements were being recorded. The considered attributes of the binary options were taste and health as represented by food images and corresponding nutritional scores, respectively. Confirming our preregistered hypotheses, participants were more likely to choose tasty over healthy food items, and this difference was amplified by hunger state. Notably, attention emerged as a pivotal mediator in this relationship. To understand the mechanism driving behavioral effects, we compared regular DDMs with attentional DDMs, multi-attribute attentional DDMs (maaDDM), and a multi-attribute time-dependent DDM that allowed the weights for taste and health to enter the choice process at different times. All models included two weights for taste and health, and versions with and without starting point bias were fitted. We found no evidence for time-dependent onsets of the two attributes. Instead, DDMs taking attention into account, specifically the maaDDM with starting point bias, best predicted our behavioral results. Together, our findings highlight the importance of taking attention into account when modeling dynamic processes of decision-making.