One of the main controversies in the field of eye movements in reading concerns the question of whether the processing of two adjacent words in reading occurs in sequence, or in parallel. This research tries to distinguish between these views by analyzing parafoveal-on-foveal effects in the course of lexical processing. Parafoveal-on-foveal (POF) effects occur when lexical properties of word n influence processing on word n－1 during reading. Different eye movement models of reading hold diverse assumption on POF effects. The E-Z Reader model, one of the best known serial attention allocation models, assumed that the POF effect are null because the attention on adjacent words is serial and the processing of word n－1 should not be influenced by the lexical processing of word n. But the parallel model SWIFT considered that the word within a perceptual span could be processed in parallel. The parafoveal processing difficulty would slow down foveal processing and the positive POF effects should be found. In the other hand, Kennedy reported an inverted parafoveal-on-foveal effect. Parafoveal processing difficulty decrease processing time on the foveal word while reading. He referred to this process as magnetic attraction. Normally, the argument to the results of POF effects focused on whether the effects were caused by saccade error. And the boundary paradigm, the most commonly used paradigm in the POF effects study, could raise saccade error and lack in balancing saccade error between experimental conditions. A new paradigm called distraction paradigm was adopted in this study to control the saccade error. The present study examined whether semantic POF effects could occur in Chinese sentence reading. 26 collage students participated in this research, average age 20.8（SD=1.6）. All participants were university students and spoke Chinese as their native language. 92 pairs of semantic related and unrelated words with word n－1 were selected. The semantic related and unrelated word n were inserted behind word n－1 directly in italic style and all participants were told to ignore them. The first pass fixation data of word n－1 were analyzed. The size of each character was 1.30cm×1.30cm (33 pixel×33 pixel). With a viewing distance of 75cm, one character space subtended approximately 0.9°of visual angle. The results showed that, compared to the related word n condition, the gaze duration of word n－1 was longer in unrelated word n condition, the refixation rate of word n－1 was also higher in unrelated condition. The semantic relationship between two adjacent words would affect the processing of word n－1 and the positive POF effects in Chinese reading were found in this study. The results supported the view that the semantic information from the parafoveal word could be processed with foveal word in parallel. And the results sustained the assumption of the typical parrel model, SWIFT model.