Pricing is regarded as a solution to congestion control in telecommunication net- works. Most mathematical models involve a so-called utility function accounting for the users’ willingness to pay. However, this utility function is unknown in prac- tice in terms of shape and important arguments. We propose here to limit this degree of uncertainty by aggregating all arguments in one quantity, the perceived quality of service, estimated using a Random Neural Network as a statistical learning tool according to the PSQA method. After arguing for this approach, we present a way of applying this tool to a model with two types of traffic and two classes of customers using strict priorities. We illustrate the proposal using a specific simple case.