Earlier identification of children with autism spectrum disorders (ASDs) is welcome, but presents a number of challenges to the clinical and the research enterprises (see Charman & Baird [2002] for a review). In the research enterprise, one critical methodological challenge is the use of appropriate measures on which to match groups of preschoolers with ASDs to comparison groups with other neurodevelopmental conditions. Language and communication impairments are central to the diagnosis of ASD and, therefore, critical variables to consider in group-matched research designs. In the domain of language function the challenges include the very poor language competence of many preschoolers with ASDs, the fact that some early language competencies form part of the formal diagnostic criteria of ASD and diagnostic algorithms on research diagnostic instruments, the uneven profile of language competency in children with ASDs, and the difference between performance on measures of formal language competency in the testing situation and everyday language use. The current paper will review these challenges and suggest some possible approaches to overcome them, including using more than one measure of language ability and adopting a pragmatic approach to group composition and statistical analysis.