We propose a computational model of human navigation, which encompasses both geometry-based and landmark-based navigation strategies. This model is based on a study of human cognitive strategies during a path memorization task in a Virtual Reality (VR) experiment. Participants were asked to memorize predeﬁned paths in a large-scale virtual city (COSMOpoliS (c)). Our computational model qualitatively reproduces the results of this experiment. This model uses the Bayesian formalism, and focuses on the interplay between the elementary cognitive strategies hypothesized above. It offers an original interpretation of the way these strategies might be articulated, departing from the classical hierarchical structure. This novel view might be fruitful for robotic models from a biomimetic perspective, where managing representations of large-scale and complex environments is still a challenge.