This paper deals with the crucial problem of comparing numerical evaluation of system performance with experimental measurements. At the moment numerical simulations permit to achieve good agreement with experiments in terms of signal evolution, but the comparison of the system performance in terms of error probability is still in progress especially when strong memory (patterning) effects are present. Advent and development of optical amplifiers have stimulated investigations of new techniques to evaluate transmission system performance. Existing operational optical systems can show no measured errors over long time intervals, that makes direct measurements of BERs almost impractical. Important role is then played by indirect methods to evaluate system performance. The most commonly used technique to evaluate system performance is Q-factor method. In its basic formulation it assumes a Gaussian noise distribution on both the zero and one levels. The optimal performance is determined by Q= (μ1-μ01+σ0). Here p1, cl0 are the means and σl,σ0 are the standard deviations of ones and zeros, respectively. BER is then calculated from Q-factor as B ER = 0.5 e rfc (Q √2) [ 11. This formulation has mainly two relatively weak points: The Gaussian hypothesis ad its unreliability when patterning effect are present.