Abstract: A blind source separation algorithm based on correntropy is presented. Unlike the traditional independent component analysis (ICA) method which utilizes the forth-order statistics or temporal structure to achieve the blind source separation. This algorithm is motivated from the notion of correntropy in the information theoretic learning, utilizing the even statistics implied in correntropy. The cost function is established according to the relationship between the parametric centered correntropy and the independence measure, and then minimized by using the optimization algorithm to acquire the demixing matrix and separate the signal. Simulations show that the performance is better than the traditional ICA method when separating the mixture of the super-Gaussian source and sub-Gaussian source.