This paper presents a method for process diagnosis based on a relational process model. To overcome the exponential complexity of the modelling and diagnostic tasks, three main results are described. First, the modelling complexity is reduced by building the overall system model from subsystem models in a component-oriented way. Second, a solution of the diagnostic problem based on relational algebra is presented and successively improved by exploiting the inherent structure of the diagnostic problem. The third idea is to decompose the diagnostic problem and to solve the resulting subproblems in parallel on a multicomputer system. A communication scheme which ensures the correct solution of the diagnostic problem is given. The theoretical results are illustrated by a diagnostic problem occurring in a power station.