This report examines how conclusions about the progress of preschool children enrolled in early intervention programmes may be influenced by the use of different cognitive assessment tools. As part of a longitudinal treatment outcome study, 24 children with autism aged between 27 and 58 months were each tested on the Bayley, Merrill-Palmer and Vineland scales. Their performance on each of these tests was compared. Results showed that, while scores on the different tests were highly correlated, actual test scores varied considerably, with the Bayley tending to produce the lowest IQ scores and the Merrill-Palmer the highest. These findings have important implications, as it is evident that judgements about the effects of therapy may be significantly influenced by the selection of the tests for pre-and post-treatment assessments. Possible ways of collecting, recording and reporting test data, so as to avoid spurious conclusions about treatment effects, are discussed.