With the current interest in using parallel computers as database servers to provide a scaleable parallel application which satisfies a real commercial need, there is a corresponding interest in performance prediction of parallel database systems. Both analytical and simulation approaches have been used and reported in the literature. This paper reports on an investigation into how a stochastic extension to classical process algebra (performance evaluation process algebra, PEPA) may be used for this purpose. This paradigm has a small but powerful set of elements which offers great flexibility for performance modelling. The paper describes how the approach has been adapted to handle database models, including the development of a technique, the decompositional approach, to handle the state-space explosion of parallel database models. It concludes with a comparison between the results obtained using this approach and those obtained using a different analytical approach.