Background: Due to limitations of individual neuroimaging methods we examine spatial and temporal contributions to self-generation using multimodality imaging with functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG) within the Bayesian framework Multiple Sparse Priors (MSP). New Method: 24 healthy participants performed an fMRI and MEG paired-associate learning task. FMRI data were processed within Group ICA fMRI Toolbox. Independent components (ICs) were temporally sorted by task time series (|r|>0.30 threshold identified task-related ICs). Task-positive (generate) ICs were retained as spatial priors for MEG analyses. MEG data were processed by averaging trials to increase the signal-to-noise ratio within subjects and with an event-related theta power approach. MEG source reconstructions were constrained within the task-positive ICs for both analytical approaches. Results: For fMRI, five networks were identified as task-related. Four ICs underlying active generation spanned bilateral parietal, orbitofrontal, medial frontal and superior temporal regions, and occipital lobe. FMRI-constrained MEG source reconstructions yielded early visual cortex activity followed by left inferior frontal gyrus (IFG) and orbito-frontal cortex (OFC) recruitment to coalesce in the left inferior temporal lobe. For the event-related theta approach, reconstructions showed a progression of activity from bilateral temporal areas to left OFC and middle temporal gyrus, followed by right IFG. Comparison with Existing Methods: MSP analyses informed by fMRI produced more focused regional activity than reconstructions without priors suggesting increased attention and maintenance when selecting relevant semantic information during active generation. Conclusions: Constraining MEG source reconstruction to fMRI priors during active generation implicates interconnected fronto-temporal and fronto-parietal networks across time.