In this paper, the author presents a methodology to build very simple but accurate enough performance models of database servers. The current available models are so complex that its utilization with real database servers is very cumbersome and not very effective. Usually, they have a large number of parameters that must be adjusted using measurements taken from real servers. The model presented in this paper is based on the queuing network paradigm using fixed distributions for the service times of the queues. The model is as simple as possible; using the minimum possible number of queues and transaction classes.