Perceiving human faces constitutes a fundamental ability of the human mind, integrating a wealth of information essential for social interactions in everyday life. Neuroimaging studies have unveiled a distributed neural network consisting of multiple brain regions in both hemispheres. Whereas the individual regions in the face perception network and the right-hemispheric dominance for face processing have been subject to intensive research, the functional integration among these regions and hemispheres has received considerably less attention. Using dynamic causal modeling (DCM) for fMRI, we analyzed the effective connectivity between the core regions in the face perception network of healthy humans to unveil the mechanisms underlying both intra- and interhemispheric integration. Our results suggest that the right-hemispheric lateralization of the network is due to an asymmetric face-specific interhemispheric recruitment at an early processing stage - that is, at the level of the occipital face area (OFA) but not the fusiform face area (FFA). As a structural correlate, we found that OFA gray matter volume was correlated with this asymmetric interhemispheric recruitment. Furthermore, exploratory analyses revealed that interhemispheric connection asymmetries were correlated with the strength of pupil constriction in response to faces, a measure with potential sensitivity to holistic (as opposed to feature-based) processing of faces. Overall, our findings thus provide a mechanistic description for lateralized processes in the core face perception network, point to a decisive role of interhemispheric integration at an early stage of face processing among bilateral OFA, and tentatively indicate a relation to individual variability in processing strategies for faces. These findings provide a promising avenue for systematic investigations of the potential role of interhemispheric integration in future studies.