Computer-aided diagnosis has become a widely-used auxiliary tool for the diagnosis of Alzheimer's disease (AD). In this study, we developed an extreme learning machine (ELM) model to discriminate between patients with AD and normal controls (NCs) using voxel-based morphometry (VBM) obtained from magnetic resonance imaging. Support vector machine (SVM), Gaussian process regression (GPR), and partial least squares (PLS) regression were compared with the ELM model. The calculated characteristics, i.e., texture features, VBM parameters, and clinical information, were adopted as the classification features. A 10-fold cross validation was used to evaluate the performance of ELM, SVM, GPR, and PLS models. We applied the proposed methods to data from 58 patients with AD and 94 NCs, and achieved a classification accuracy of up to 0.96 with all classification features of the ELM model, while the results of the other three models were 0.82 (PLS), 0.79 (GPR), and 0.75 (SVM). Furthermore, the effect of VBM parameter modeling is better than texture parameter. Thus, our method was optimal in distinguishing patients with AD from NCs, and may therefore be useful for the diagnosis of AD.