Abstarct: Deep learning is a new branch of machine learning, which is shown promising result compared to other learning algorithms. Deep learning algorithm are capable to represent and classify a set of data in hence of their hierarchical structure. Moreover, deep structures provides comprehensive presentation in comparison with shallow structures. Deep learning is being used in machine vision, voice recognition, natural language processing and signal processing. In this work, we used Deep learning to classify EEG signals representing different class of emotions. Emotional states influence human’s daily life. Detection and recognition of emotional states is necessary in human-machine interaction in order to make these interactions more similar to human-human interaction. In the proposed method, EEG signals are decomposed by empirical mode decomposition (EMD). EMD decomposes EEG signal into a set of intrinsic mode functions (IMFs). Then singular value decomposition (SVD) is applied to extracted IMFs. IMFs’ components corresponded to higher eigen values are selected and the others are discarded. IMFs are reconstructed using components related to higher Eigen values. IMFs’ spectrum is obtained using Hilbert transform. In feature extraction step, spectral moment of IMFs is calculated as features. Finally, a type of Deep belief network composed of Restricted Boltzmann Machine is being used to classify EEG signal to detect emotion in 2 dimensional model of valence and arousal. In this research DEAP databaxse is used to evaluate the proposed method. baxsed on the obtained results, it is seen that the proposed method performs better than other algorithms applied on DEAP databaxse. The result of classification of 3 classes of valence and arousal are 60% and 68% respectively.