AUTISTIC children have been shown to be able to form an object discrimination learning set and to be superior in their performance to normal and retarded mental age matched control groups (Prior and Chen, 1975). It has been suggested that this superior performance in a diagnostic group whose cognitive performance is usually found to be at a very low level (e.g. De Myer el al., 1974) may be due to the fact that verbal mediation is not essential for the formation of a learning set and/or the fact that the object discrimination task employed in the earlier experiment was an extremely simple one. The experiment reported here was designed to extend the findings of that experiment both in terms of an increase in difficulty of the learning task and in terms of assessing the ability for more abstract cognitive functioning known to be deficient in autistic children (Kanner, 1943;Hermelin and O'Connor, 1970). The experimental paradigm employed was that of the conditional matching learning set (CMLS). The basis for this paradigm was devised by Harlow (1958) for use with primates and was called in this context, the combined oddity-non-oddity problem. Hill (1965) using the Harlow type conditional oddity problem (where the rewarded response was either the odd or the non-odd stimulus depending on the colour of the tray on which the stimuli were presented), reported that only three of 30 normal 6-yr-olds could solve this problem in 200 trials.The CMLS has been adapted from the Harlow paradigm and used as a technique for studying the abilities of children with communication disorders by Levin, Rapin, Costa and Tourk (1971) who have suggested that it may be a better way of assessing abilities of such children than standard psychometric tests. Levin et al. (1971) showed that both normal children and children with communication disorders of a mental age of 6-14 years could solve the CML Sproblems even though for some cases LQ,. was in the moderately retarded range and language impairment was severe. These authors also found that CMLS ability for both groups was reliably related to M.A. and to LQ,.and that it rehably predicted academic performance one year later. Since their handicapped subjects seem to have been relatively more successful than Hill's normal 6 yr-olds it might be suggested that the CMLS is not as difficult as the conditional oddity problem or that there are additional cues available in the CMLS task to assist in its solution. It should also be noted that Levin et al. used a pre-training procedure to facilitate learning. It seemed then that although the CMLS was a difficult task for young children, it •Requests for reprints to: