AbstractAn industrial afterburner was optimized using statistical experimental design. A computer‐generated optimal design was adapted to domain constraints and previous knowledge about the experimental variables. Carbon monoxide was used as a general indicator for the operating conditions, and a reduced response surface model was fitted to the experimental data. In a second phase the initial results were verified and investigated in more detail with regard to chlorinated aromatics. The empirical models obtained were further interpreted using results from laboratory investigations and more fundamental knowledge from combustion theory. Copyright © 2003 John Wiley & Sons, Ltd.