In this paper we evaluate the performance of 5 source separation algorithms (AMUSE, SOBI, SOBI-RO, SONS, JADE-TD) on simulated EEG signals. A first result evaluates the influence of the noise and signal characteristics (frequency, length, SNR) on the algorithms performance. A second objective is to introduce a new performance criterion, Eigen Values Vector's Norm-1 Distance (IEV) which can be used to compare two matrices and is potentially useful on real signals. We validate this new index by comparing it with classic performance indices used in source separation.