The question of when it is best to screen for autism may only be answered by a series of empirical studies. These will be difficult to plan, fund, and conduct, and will by necessity take many years because of the need to systematically follow up the whole cohort screened. In our study, we identified 19 of the 50 children with autism by their profile at the 18-month screen (though note that some fell out of risk status at the repeat screen 1 month later--thus sacrificing sensitivity for improved positive predictive power). Through the subsequent surveillance methods we employed, we identified the remaining cases as follows: 5 at 42 months, 4 between 42 months and 7 years, and 25 at 7 years. We do not mean to end on a pessimistic note. Our experiences have been positive both in regard to the instrument we developed and the effects that using it have had on the health practitioners involved in the research study. In discussion, practitioners have commented on the usefulness of knowing what prelanguage and prosocial skills can reliably be looked at during the 18-month check. Training using the CHAT and eliciting its behaviors improved the skills and confidence of primary health practitioners. It is our view that this has had the effect of reducing the age at which autism is recognized and cases are referred on for a developmental assessment. The work reported by Robins er al. makes an important contribution to this ongoing research and clinical process as we attempt to accurately identify children with autism at a young age.