The present research aimed to investigate children’s comprehension of Chinese classifiers. Sixty-five Chinese-speaking children between the ages of 4 and 6 recruited in Taiwan participated in the experiment. The results indicate that children can make generalization based on their understanding of classifiers instead of solely relying on classifier-noun associations. The results also show that the participants performed equally on both shape-based and feature-shared classifiers, which suggests that children not only use shape salience to learn Chinese classifiers, but are also sensitive to other relations between objects classified by the same Chinese classifier. Besides, the complex patterns in the results imply that in spite of the exposure to classifiers, the semantic transparency between classifiers and objects varies considerably in both semantic types of classifier, which might be the primary reason that some classifiers are more difficult for children to acquire.