Markov arrival processes (MAPs) are used extensively in traffic modeling. Consequently a wide variety of fitting procedures have been developed. Most of these however are computationally demanding or not general enough. To resolve this problem, two-step procedures of a specific type have been made, which fit a phase-type distribution (PH) to static parameters in the first step, and extend it to a MAP in the second while fitting dynamic parameters. Their general weakness is that the first step often restricts the attainable range of dynamic parameters. In our paper we present a method, that aims at providing a good starting point for the second step, by optimizing the representation of the PH that was produced by the first step.