A robust air/fuel ratio "soft sensor" is presented based on non-linear signal processing of the ion current signal using neural networks. Care is taken to make the system insensitive to amplitude variations, due to e.g. fuel additives, by suitable preprocessing of the signal. The algorithm estimates the air/fuel ratio to within 1.2% from the correct value, defined by a universal exhaust gas oxygen (UEGO) sensor, when tested on steady state test-bench data and using the raw ion current signal. Normalizing the ion current increases robustness but also increases the error by a factor of two. The neural network soft sensor is about 20 times better in the case where the ion current is not normalized, compared with a linear model. On normalized ion currents the neural network model is about 4 times better than the corresponding linear model. Copyright © 1999 Society of Automotive Engineers, Inc.