To underscore the increasing importance of college studentsability in spoken English,the National College English Testing Committee began to administer the College English Test-Spoken English Test(CET-SET)in 1999.One distinctive part of the CET-SET is a small group discussion among candidates themselves which intends to engage candidates in interactive communication and thus measure their interactive speaking ability.However,the likely discrepancy between the test designersintentions and the candidates actual performance calls for the need for demonstrating construct validity,that is,to what extent the elicited test performance reflects the areas of speaking ability the test designers intend to assess.Unfortunately,there is little research to date in this regard.The present paper undertakes a corpus-based study of the construct validity of the CET-SET group discussion by examining the match between intended and actual performance in the group discussion task with respect to a set of Interactional Language Functions to be assessed.The results show a low degree of interactiveness between candidates in the CET-SET group discussion.Consequently,the inadequate elicitation of Interactional Language Functions from candidates poses the problem for measuring their speaking ability in this regard.