Residence time distribution (RTD) measurements can be interpreted and simulated using compartment models which consist in interconnection of elementary modules, e.g. the plug flow reactor or the perfect mixing cell. However, the choice of the appropriate model may be a troublesome task when the considered reactor is of a complex geometry. The present investigation was aimed at developing a new algorithm for the automatic generation of compartment models based both on RTD curves and the physical description of the studied processes.We used for this purpose the possibility theory developed by Zadeh (Fuzzy Sets and Systems 1 ( 1978) 3), which consists in affecting weighing coefficients on the possible occurrence of a given event. The use of this method for chemical engineering has been proposed by Kraslawki (Chem. Eng. Process. 26 (1989) 185) but so far, it has been mainly tested in human sciences, as by Sandri and Bittercourt (1994) for investigations of associative human memories. In this context the compilation of existing work allowed the selection of nearly 40 models, which makes it possible to describe a large variety of industrial situations. Based on the RTD curves and process characteristics, a list of rules could be deduced. Weighing coefficients for possibility degree and necessity degree have been introduced to link the rules of the models. Finally, the so-called "fusion" procedure consisting of the compilation of all rules, leads to the selection of the most appropriate model. Addition of new rules or new models affects neither the structure of the software nor the "fusion" procedure. The software is very flexible and its evolutions are only limited by the user's expertise in chemical engineering. This paper deals mainly with the description of the algorithm and few examples of illustration. (C) 2003 Elsevier Ltd. All rights reserved.