Many graph-based studies consider just one habitat type (breeding habitat) and one ecological process (dispersal) when measuring connectivity. However, in the course of their life cycle, some species use different habitats and are particularly sensitive to their spatial arrangement. We propose a new graph modeling approach that considers heterogeneous habitats and movements when assessing connectivity. Our hypothesis is that a multiple habitat approach is more relevant than a single habitat approach for studying the relationships between connectivity and composite habitat species occurrence. We constructed a bipartite graph with separate categories of nodes for aquatic and terrestrial habitats and an inter-habitat link set. Different connectivity values were measured from the graph depending on the ecological process under study. We then compared our models against field observation data for amphibian communities and tested whether multiple habitat connectivity models provide better predictions. Multiple habitat connectivity was as effective as single habitat connectivity, and often more so, in explaining the presence, abundance, and species richness of amphibian communities. Application to habitat restoration revealed that connectivity gains vary greatly depending on the ecological process, emphasizing the importance of links connecting habitats of different types. Multiple habitat graphs appear promising for incorporating habitat and movement heterogeneity into ecological network modeling. We encourage further theoretical and empirical work on multiple habitat connectivity to consolidate the results and evaluate its full potential, especially for global change issues.