Processing of facial information is distributed across several brain regions, as has been shown recently in many neuroimaging studies. Disturbances in accurate face processing have been repeatedly demonstrated in different stages of schizophrenia. Recently, electroencephalography (EEG) and tomographic analysis of average magnetoencephalographic (MEG) data were used to define the latencies of significant regional brain activations in healthy and schizophrenic subjects elicited during the recognition of facial expression of emotions. The current study re-examines these results using tomographic analysis of single trial MEG data. In addition to the areas identified by the analysis of the average MEG data, statistically significant activity is identified in several other areas, including a sustained increase in the right amygdala activity in response to emotional faces in schizophrenic subjects. The single trial analysis demonstrated that the reduced activations identified from the average MEG signal of schizophrenic subjects is due to high variability across single trials rather than reduced activity in each single trial. In control subjects, direct measures of linkage demonstrate distinct stages of processing of emotional faces within well-defined network of brain regions. Activity in each node of the network, confined to 30 to 40 ms latency windows, is linked to earlier and later activations of the other nodes of the network. In schizophrenic subjects, no such well-defined stages of processing were observed. Instead, the activations, although strong were poorly linked to each other, managing only isolated links between pairs of areas.