The Revised Behaviour Summarized Evaluation Scale (BSE-R) was developed for the objective evaluation of autistic behaviours in order to facilitate the recording of the evolution of developmentally disabled children. Among its 29 items, 13 items that precisely describe the degree of autistic behaviours were extracted by Principal Component Analysis. We hypothesised that these relevant behaviours could differentiate autistic behaviour profiles in a population of children previously diagnosed as typically autistic. For this purpose, we used an original multivariate descriptive statistical approach, Correspondence Analysis, which can help in detecting structural relationships among variables. In a population of autistic children initially diagnosed using DSM-III-R criteria, this procedure proved effective in identifying new main dimensions of behaviours among the 13 previously defined core autistic symptoms. Cluster analysis, which followed factorial analysis, allowed the identification of three meaningful behaviour profiles separated principally on the basis of two main functions, perception and imitation, which have been always considered to be fundamentally involved in autistic syndrome. The individualisation of homogeneous subgroups of children on the basis of the behavioural evaluation provides a potentially useful starting point for further biological and therapeutic studies.