Efficient monitoring and automatic control systems for biological wastewater treatment processes, especially those dealing with bioinhibitory pollutants, such as phenol, are urgently required in order to meet increasingly stringent environmental regulations. Practical on-line sensors of variables that describe water quality, such as BOD or individual toxic pollutants such as phenol, are not commercially available; e.g. phenol is generally monitored off-line by spectrofluorometry. Inference software sensors could be an attractive alternative for on-line monitoring of these variables. As a first step towards the development of inferential sensors for biological wastewater treatment processes, we consider in this study, a simplistic version of such a process which consists of a continuous culture of Pseudomonas putida Q5 degrading phenol. In this work, we propose a neural network based inferential sensor for phenol monitoring using on-line biomass concentration measurements by spectrophotometry. The network was built with wavelets as the basis functions and the adaptive algorithm for the weights was based on a Lyapunov stability analysis. Predicted phenol output of the network showed good agreement with experimental data, over fairly broad ranges of inlet phenol concentration and dilution rate step changes. Simulations were conducted to find convergence conditions and to investigate possible sources for errors in phenol estimates.