Masked repetition has been suggested as the reliable paradigm investigating the pre-lexical processing of English words. This suggestion has the supportive evidence that masked repetition effects do not interact with word frequency. Chinese compound words are the combinations of two constituent characters which have independent lexical properties. The lemma model argues that the lemmas of constituents would increase masked repetition effects of low-frequency compound word when the constituent characters have relative higher frequency. To verify this argument, this study manipulated the frequency and the morphological aspects of constituent characters. The critical results are the increased masked repetition effect of low-frequency coordinative words which have at least one high-frequency characters and the null interaction of masked repetition effect and word frequency for the compound words having meaningless constituents. The approaches to modify the current lemma model are discussed according to these findings.