Neuropsychiatric disorders such as autism spectrum disorder (ASD), schizophrenia (SCZ), and major depressive disorder (MDD) affect individuals at different stages in life. Current diagnostic procedures for these disorders rely heavily on behavioral criteria, making diagnosis difficult. Because treatment options are based on correctly identifying the disorder, patient assessment based on more objective measures becomes essential. One potential objective biomarker to diagnose disorders is called brain age prediction. Using imaging data, the brain is predicted and then compared to the actual biological age of the subject. The difference between the two, called brain age gap (BAG), is then used as a biomarker for the disorder. In this study, we examined the effectiveness of BAG as a diagnostic tool to classify patients with ASD, SCZ, and MDD from healthy controls. We further investigated different preprocessing methods and prediction models to improve the accuracy of the classification. MRI data from a publicly available repository were used in the analysis. From this pool of participants, we extracted structural MRI from healthy controls (HC, N = 791) and patients with ASD (N = 125), SCZ (N = 147), and MDD (N = 255). To predict brain age, we used three regression models including linear and quadratic support vector machines for regression (SVR) as well as LASSO. Our results showed that ASD and SCZ have significantly (p < 0.001) older predicted brain age (higher BAG values) compared to HC. By contrast, MDD appeared to show young brain age. ROC analyses further suggest that ASD and SCZ, but not MDD, can be classified from HC using BAG values with accuracy greater than chance. Across the different predictive models, LASSO provided the best prediction for brain age with the lowest mean absolute error value. However, this does not translate to better classification performance of the different disorders. In fact, LASSO’s performance was the worst of the three models. By contrast, linear SVR provided the best classification performance. Taken together, these findings suggest the potential use of BAG, estimated using structural MRI data, as a neuroimaging biomarker for the classification of ASD and SCZ, but not MDD, from healthy controls.