This paper deals with the design and the application of Artificial Neural Networks (ANN) to the fault accommodation of the mass air flow meter in modern diesel engines. Several ANN architectures are proposed and tested. In order to verify their performance in terms of accuracy and promptness, a typical graphical tool (regression error characteristic curves) and an original one proposed by the authors (named sliding occurrence error curves) are applied. In addition, to prove the real applicability of the proposed ANN architectures for on-line implementations, suitable computational burden indexes are evaluated to quantify the processing resource requirements and to verify their compatibility with typical microprocessor adopted in the automotive context. A large experimental campaign has been performed for two of the most widespread diesel technologies, common rail and injection pump. The achieved results show: i) the suitability of the proposed graphical tools in evaluating and comparing the ANN performance; ii) the good performance of the proposed architectures in terms of accuracy, promptness, and computational burden.