Magnetic resonance imaging (MRI) data has been instrumental in the identification of structural changes associated with multiple psychiatric disorders. Thus, several studies have explored the use of MRI data as an objective measure for the classification of different psychiatric disorders. Many of these studies typically used imaging data acquired from the same site both for training a given classifier and testing its performance. However, MRI data have features that differs from site to site due to differences in MRI scanners, imaging protocols, and clinical characteristics of the patient group, which could influence the performance of trained classifiers. Using publicly available MRI datasets, this study aims to classify patients with diverse neuropsychiatric disorders including major depressive disorder (MDD), autism spectrum disorder (ASD), and schizophrenia (SCZ) from healthy controls (HC) based on structural MRI data acquired using different MRI scanners with different imaging parameters located at different imaging sites using support vector machine (SVM). In addition, we also aim to identify patterns of structural brain changes that contribute significantly to the classification, which may serve as potential imaging-based biomarkers for the different disorders. For two-class classifications, obtained accuracies ranged from 50.8 to 59.5% for HC vs MDD, 52.7 to 65.4% for HC vs SCZ, and 52.1 to 60.0% for HC vs ASD depending on the approach used to preprocess the data and the sites where data used for training and testing were acquired. For multi-disorder classification, the accuracy ranged from 34.1 to 55.9%. The regions contributing to the classification of patients with MDD included the cerebellum, precentral gyrus, and cuneus, those with SCZ included the hippocampus, thalamus, frontal gyrus, and cerebellum, and those with ASD included the insula, precuneus, and cuneus. These regions have been reported to be related to the pathogenesis of each disorder.