Complexity metrics are widely used to estimate the difficulty of understanding process models. However, the relationship between these metrics and the concept of cognitive load, which captures the difficulty experienced by users, is not fully understood in the process modeling literature. In neighboring fields like Software Engineering, researchers could only to a limited degree establish a relationship between complexity metrics and users’ cognitive load. To investigate the extent to which such a relationship exists in the process modeling field, we conduct an eye-tracking experiment that assesses how a suite of metrics, capturing both the essential complexity inherent to the process specifications and the accidental complexity emerging from the model layout, aligns with users’ cognitive load during model comprehension tasks. Our findings show that the used metrics suite aligns well with users’ cognitive load. Moreover, our analysis of users’ behavior suggests that different levels of model complexity yield distinct visual behaviors. The implications of our work extend to both practice and research, validating a comprehensive suite of complexity metrics and delivering a multi-granular approach that can be reproduced in other experiments to enable the analysis of users’ cognitive load and behavior on simple but also complex models.