Previous studies on Sino‒Japanese-derived words have focused on the results of decomposition into the base identity and affix. However, the time course of lexical access has not yet been investigated. There is a possibility that, before the decomposition process occurs, lexical access is initiated through the use of clues from the initial characters. To investigate the time course of lexical access, we used a masked priming technique where we used a derived word as a prime to examine the reaction time upon presenting the base word as a target. Because the stimulus onset asynchrony (SOA) was 240 ms, the early stage of lexical access could be investigated. The results revealed that a derived word is not necessarily decomposed. However, some derived words seem to be decomposed, and this depends on the word’s construction. This also suggests that the early stage of lexical access occurs with the priority of processing the initial character, not with the priority of processing the base identity.