Large-scale networks support the dynamic integration of information across multiple functionally specialized brain regions. Network analyses of haemodynamic modulations have revealed such functional brain networks that show high consistency across subjects and different cognitive states. However, the relationship between the slowly fluctuating haemodynamic responses and the underlying neural mechanisms is not well understood. Resting state studies have revealed spatial similarities in the estimated network hub locations derived using haemodynamic and electrophysiological recordings, suggesting a direct neural basis for the widely described functional magnetic resonance imaging (fMRI) resting state networks. To truly understand the nature of the relationship between electrophysiology and haemodynamics it is important to move away from a task absent state and to establish if such networks are differentially modulated by cognitive processing. The present parallel fMRI and magnetoencephalography (MEG) experiment investigated the structural similarities between haemodynamic networks and their electrophysiological counterparts when either the stimulus or the task was varied. Connectivity patterns underlying action vs. object naming (task-driven modulations), and action vs. object images (stimulus-driven modulations) were identified in a data driven all-to-all connectivity analysis, with cross spectral coherence adopted as a metric of functional connectivity in both MEG and fMRI. We observed a striking difference in functional connectivity between conditions. The spectral profiles of the frequency-specific network similarity differed significantly for the task-driven vs. stimulus-driven connectivity modulations. While the greatest similarity between MEG and fMRI derived networks was observed at neural frequencies below 30 Hz, haemodynamic network interactions could not be attributed to a single frequency band. Instead, the entire spectral profile should be taken into account when assessing the correspondence between MEG and fMRI networks. Task-driven network hubs, evident in both MEG and fMRI, were found in cortical regions previously associated with language processing, including the posterior temporal cortex and the inferior frontal cortex. Network hubs related to stimulus-driven modulations, however, were found in regions related to object recognition and visual processing, including the lateral occipital cortex. Overall, the results depict a shift in network structure when moving from a task dependent modulation to a stimulus dependent modulation, revealing a reorganization of large-scale functional connectivity during task performance.