Two parallel experiments were designed to examine how contextual diversity modulated the processing of initial and final character’s positional probability in Chinese word identification. In Experiment 1, we manipulated the initial and final characters’ positional frequencies for two-character words with high contextual diversity. In Experiment 2, an analogous manipulation was made for low contextual diversity words. Sixty participants were instructed to take part in the lexical decision task in the two experiments. The results showed that, for the identification of high contextual diversity words, Chinese readers were sensitive to word’s final character positional frequency only when the word’s initial character was not frequently occurring at word beginning. In contrast, both the initial and final character’s positional frequency were processed for the identification of low contextual diversity words. We argue that these findings support the augmented addressed morphology model theory of Chinese word recognition.