This article presents a queueing model for performance evaluation of a large database system at an assurance company. The system includes a server with a database and a local area network with a number of terminals where the company employees run applications that introduce documents or retrieve them from the database. We discuss a model of clients activities. Measurements were collected inside the working system: the phases at each users application performance were identified and their duration was measured. The collected data are used to construct a synthetic model of applications activities which is then applied to predict the system behaviour in case of the growth of the number of users. We apply simulation, Markov and diffusion models-their comparison, based on real data, may better verify the utility of particular methods than usual academic examples.