In this paper, we propose a new approach for automatic musical genre classification which relies on learning a feature hierarchy with a deep learning architecture over hand-crafted feature extracted from the audio signal. Unlike the state-of-the-art approaches, our scheme applies an unsupervised learning algorithm based on Deep Belief Networks (DBN) on block-wise MFCC (that we treat as 2D images) followed by a supervised learning algorithm for fine-tuning using the labels for the music. Experiments performed on the GTZAN dataset show that the proposed scheme clearly outperforms the state-of-the-art approaches.