This paper addresses the problem of finding the decisive parameters, the so-called key parameters, of a queueing system which most significantly influence expected occupation and loss of a finite capacity queue. It is shown that the key parameters of observational data, which are the expected arrival rate, the expected service rate and the spectral density at the frequency zero of the difference between arrivals and service time most significantly influence the performance of the queueing system. An algorithm is developed which shows that it is mostly possible to fit an MMPP(2) to the key parameters. A numerical example illustrates the importance of the key parameters and also shows the accuracy of the proposed fitting procedure.