This work intends to studying and simulating the cognition of preschool children. Two domains were specifically studied, because of the rich scope of young children's vocabulary in these areas: food and animals. Free sorting task was applied on a group of preschool children and salient dimensions (biological, perceptual) in the process of food and animals categorization were selected from their verbalizations. Data were analyzed using Kohonen maps and fuzzy clustering. These two complementary approaches prove effective to address the complex task of clustering, to allow taxonomy enrichment and to predict children reactions to a new animal or food. The perspectives of extending this experimentation are rich, both in the field of AI and that of psycho or socio-linguistics: compared cognitive developments, study of stereotypes, etc.