The Infomax algorithm is a popular method in blind source separation problem. In this article an extension of the Infomax algorithm is proposed that is able to separate mixed signals with any sub- or super-Gaussian distributions. This ability is the results of using two different nonlinear functions and new coefficients in the learning rule. In this paper we show how we can use the distribution of observation vectors for selecting suitable coefficients for our algorithm. Hence, the proposed algorithm is suitable for real applications in which the distribution of source signals might be unknown. It is also shown in this paper that the extended Infomax algorithm is able to separate 23 sources with a variety of distributions. Incidentally, we use a performance criterion for the evaluation of our results, based on the comparison of Kurtosis of the original signals and estimated signals.