Fast detection of abnormal plant operation is critical for many applications. Fault detection requires some kind of comparison between actual and normal behavior, which implies the use of models. Exact modeling of engine systems is probably impossible and even middle-complexity models is very timeconsuming. In some few cases, as for on board diagnostics, the very limited amount of cases to be treated and the usually large production volumes allow to develop models suitable to detect an abnormal behavior, but, in general, however, this approach cannot be followed. As fast detection of abnormal plant operation is often critical, alternative low-effort approaches are required. This paper presents a procedure suitable for engine fault detection based on parallel automatic modeling. It is shown that this approach yields a flexible and reliable tool for automatic modeling for this goal, while keeping the effort for the operator rather low.