The effect of children's semantic knowledge on the class inclusion task was examined in 2 experiments. In Experiment 1, questions with typical subset exemplars were easier for 6 year old children than questions with atypical exemplars, and also easier than questions with geometrical figures. The typicality of the exemplars was especially effective among the Ss who could succeed in some of the geometrical questions. In Experiment 2, questions with super-ordinate class labels which were more easily produced from subset exemplars were observed to be easier in most of the cases. The effect of the development in children's semantic knowledge on the process of acquiring the ability to solve the class inclusion task was further discussed.