The issue addressed in this paper is the determination of the number of communication signals in a sensor array. Most of the available algorithms rely on the spatial uncorrelation of the additive noise. In practice, this condition is rarely satisfied when the receivers are not sufficiently spaced (MIMO communications for example). In this paper, we propose a new method to detect the number of communication signals based on the fact that the signals are independent and non gaussian and that the background noise is gaussian. By using an Independent Component Analysis in conjunction with Kolomogorov-Smirnov (K-S) tests, the method can detect as many communication signals as the number of receiver antennas. Simulations results show that our method performs well in many environments like those with spatially correlated noise.