In high-tech production, companies often deal with the manufacture of assemblies with quality requirements close to the technological limits of manufacturing processes. The article shows an approach of a virtual in-line inspection, predicting the products functionality. An artificial neural network (ANN) fed with product characteristics and process data as well as the resulting functional fulfillment of the product is trained for virtual function prognosis. Through the preventive identification of defective products before the final assembly step, components can be recovered and returned to serial production. By optimizing the parameters of the ANN, incorrect classifications are reduced and the practical applicability is ensured. The approach is demonstrated in an industrial application in the automotive industry.