The aim of this work is to understand the role of the micro-mixing model in computational fluid dynamics (CFD) simulations of fast reactions. Using CFD, in fact, the reactor is modelled through a computational grid and the governing equations are discretised using numerical methods. However, the mixing phenomena that occur at scales that are smaller than the grid size remain unresolved. This means that the probability density function (PDF) of all scalars is assumed to be a delta function centred at the mean value. In order to take into account micro-mixing effects a model must be added. In this work the finite-mode PDF approach is used to predict the selectivity of a parallel reaction in a Taylor-Couette reactor working in semi-batch conditions. Experimental data are compared with model predictions in order to investigate the relevance of the micro-mixing model. The case of precipitation is also discussed. (C) 2003 Elsevier Ltd. All rights reserved.