Recently, the early diagnosis of Alzheimer’s disease (AD) has gained major attention due to its growing prevalence and the resulting costs imposed on individuals and society. While deep learning (DL) techniques have made significant progress in medical image analysis, their application to AD diagnosis has received relatively less attention. Moreover, although these techniques in the AD diagnosis are very prevalent, current performances are still not robust or satisfactory. The main goal of this study is to find out how well other DL techniques, like the more recent convolutional neural networks (CNNs) architectures, can be used to diagnose AD at an early stage. We design a very deep convolutional network by introducing the Place365-CNNs architecture, which is pre-trained using a large Places database. With its high coverage and high diversity of exemplars, the Places database is along with CNNs to improve the diagnosis performance for the 4-way classification of AD. Using seven performance metrics, tests on brain MRI data from the Open Access Series of Imaging Studies dataset show the effectiveness of the proposed model. Furthermore, we compare the obtained results with state-of-the-art transfer learning approaches and some published works, demonstrating their superior performance over the other methods.