Nowadays, firms, formerly considering the hu- man operator as the main error source in pro- cess control, bend their efforts towards anthro- pocentric approaches to (re)integrate the hu- man factor, especially the knowledge he/she has been developping, as the essential resource for a high quality decision process. As the expert operator remains a rare resource and in order to capitalize his/her knowledge and know-how, the development, of tools inte- grating this new dimension has become an im- portant challenge. This paper deals with a tool for knowledge ac- quisition under cognitive constraints, assuming that cognitive principles could be sometimes useful to improve machine learning tools re- sults. Additionally, we have to cope with the difficulty linked to the fact that the acquired strategies have to be adapted on-line. After describing the underlying cognitive prin- ciples, we will introduce the decision represen- tation space and its related notations. We will then show the difficulties linked to the search of an optimal representation of the expert strate- gies set and how the heuristics used by the al- gorithm studied avoid these NP-complete prob- lems. Finally, the current results and our work perspectives are stated.