There is strong evidence to suggest that the complex cognitive process underlying mental rotation does not have a discrete neural correlate, but is represented as a distributed neural system. Although the neuroanatomical nodes of this so-called rotation network are well established, there is as yet little empirical evidence to indicate how these nodes interact during task performance. Using an optimized, event-related paradigm, this study aimed to test a previously proposed hypothetical neurocognitive network for mental rotation in female subjects with path analysis, and to examine changes in effective connections across different levels of task difficulty. Path analysis was carried out in combination with a time-resolved functional magnetic resonance imaging (fMRI) analysis in order to relate the observed changes on the network level to changes in specific temporal characteristics of the hemodynamic response function on the level of individual neuroanatomical nodes. Overall, it was found that the investigated sequential model did not provide an adequate fit to the data and that a model with parallel information processing was superior to the serial model. This finding challenges traditional cognitive models describing the complex cognitive process underlying mental rotation by a set of sequentially organized, functionally distinct processing stages. It was further demonstrated that the observed in interregional effective connectivity changes with the level of task demand. These changes were directly related to the time course of the experimental paradigm. The results of path analysis in fMRI should therefore only be interpreted in the light of a specific experimental design and should not be considered as general indicators of effective connections.