Travel time prediction is a challenging problem in Intelligent Transportation Systems (ITS). Accurate travel time information helps motorists plan their routes more wisely. This, in turn, alleviates traffic congestion and improves operation efficiency. A number of travel time prediction techniques exist; however, most of them are based on shallow learning architectures. In contrast to deep learning architectures, shallow learning architectures are lack of features-learning capability. In this paper, we propose an effective travel time prediction technique based on a concept of Deep Belief Networks (DBN). In our method, a stack of Restricted Boltzmann Machines (RBM) is used to automatically learn generic traffic features in an unsupervised fashion, and then a sigmoid regression is used to predict travel time in a supervised fashion. The experimental results, based on real traffic data, show that the proposed method can achieve great performance in terms of prediction accuracy.