The present study aimed to explore altered effective connectivity in schizophrenic patients while performing a 2-back working memory task. Twelve right-handed, schizophrenic patients treated with typical or atypical antipsychotics and 6 healthy control subjects were studied with fMRI while performing a "2-back" working memory task. Effective connectivity within a cortical-subcortical-cerebellar network for mnemonic information processing was assessed and compared between both groups. The path model included cortico-cortical connections comprising the parietal association cortex, ventrolateral prefrontal cortex (VLPFC), and the dorsolateral prefrontal cortex (DLPFC) as well as a cortico-cerebellar feedback loop comprising prefrontal cortex, contralateral cerebellum, and thalamus. Group differences were analyzed with a stacked models approach. Relative to normal controls, both patient groups revealed a pattern of reduced connectivity within the prefrontal-cerebellar and the cerebellar-thalamic limbs but enhanced connectivity in the thalamo-cortical limb of the cortical-cerebellar circuit. Moreover, a direct comparison of both treatment groups revealed enhanced connectivity in the interhemispheric connections between the cortical association areas in patients treated with atypical antipsychotics. However, right prefrontal and left parieto-frontal path coefficients were lower in the patient group receiving atypical antispychotic drugs. The findings suggest that the relationship between pathology in cortical-subcortical cerebellar networks and associated functional connectivity is complex and may include aspects of increased and decreased levels of connectivity consistent with the notion of "cognitive dysmetria" in schizophrenia. The observed pronounced connectivity within thalamo-cortical projections could be attributed to a compensatory increase of thalamic input in the presence of disrupted effective connectivity within the preceding limb of the cortical-cerebellar circuitry. The study demonstrated the feasibility of structural equation modeling for the investigation of group and treatment-related differences in effective connectivity and provides a promising approach to further disentangle the relationship between altered functional capacity and associated fMRI signal changes.