A trial-based spatial working memory (WM) paradigm, where trial type was unpredictable at encoding, was used to evaluate differential predictions of various models of SWM organization in humans (Study 1) and disruptions in this network in patients with schizophrenia (Study 2). Study 1 revealed that both maintenance (M) and manipulation (MnM) trials were associated with activation of frontal, parietal, occipital and thalamic regions, with significantly greater activation for MnM than for M. In both trial types, occipital cortex and FEF were more active in the early phase while the left frontal pole and VLPFC were more active in the later phase. These results suggest a relatively greater role of DLPFC, FEF, premotor and thalamus in MnM of a maintained representation, and a relatively greater role of VLPFC in comparison operations and response selection. Neurophysiological models that may account for a differential deficit in MnM include those that specify a locus of dysfunction in DLPFC and those that specify network or circuit-based dysfunction involving prefrontal interactions with medial temporal lobe or thalamic and cerebellar regions. Results indicated a marginal difference between patients with schizophrenia and controls on MnM accuracy, but no group differences in reaction time. Compared with performance on a prior blocked version of the task in which trial type was predicable at encoding, performance on M trials was more significantly degraded than on MnM trials, and to a much greater degree in controls than in patients, suggesting that patients fail to establish flexible WM representations conducive to MnM. Patients showed significantly less activity than controls in the VLPFC and hippocampus. For all other regions, group differences in activation were specific to trial type, trial phase, and/or hemisphere. Patients showed relatively greater activity than controls on MnM compared with M trials in FEF, ACC, PSPC, and premotor cortex, an effect that was accounted for by significantly lower activity in patients than controls on M trials. Overall, the results are in line with network wide dysfunction in the fronto-temporal and cortico-thalamic circuits and suggest deficits in both encoding and retrieval phase WM processing, with patients employing a more literal, storage-biased encoding strategy and controls employing a higher effort, flexible encoding strategy more conducive to MnM operations.