The present study examined the use of statistical cues for word boundaries during Chinese reading. Participants were instructed to read sentences for comprehension with their eye movements being recorded. A two-character target word was embedded in each sentence. The contrast between the probabilities of the ending character (C2) of the target word (C12) being used as word beginning and ending in all words containing it was manipulated. In addition, by using the boundary paradigm, parafoveal overlapping ambiguity in the string C123 was manipulated with three types of preview of the character C3, which was a single-character word in the identical condition. During preview, the combination of C23′ was a legal word in the ambiguous condition and was not a word in the control condition. Significant probability and preview effects were observed. In the low-probability condition, inconsistency in the frequent within-word position (word beginning) and the present position (word ending) lengthened gaze durations and increased refixation rate on the target word. Although benefits from the identical previews were apparent, effects of overlapping ambiguity were negligible. The results suggest that the probability of within-word positions had an influence during character-to-word assignment, which was mainly verified during foveal processing. Thus, the overlapping ambiguity between parafoveal words did not interfere with reading. Further investigation is necessary to examine whether current computational models of eye movement control should incorporate statistical cues for word boundaries together with other linguistic factors in their word processing system to account for Chinese reading.