Rapid automatized naming (RAN) deficit is one of the cognitive deficits of dyslexic in both alphabetic languages and Chinese. Two hypotheses have been put forward to account for the RAN dysfunction in dyslexic: reduced parafoveal preview benefits and increased parafoveal load costs. At present, these two hypotheses have not been distinguished because previous studies only manipulated one of the two variables in one experiment; however, the manipulations of parafoveal preview and parafoveal load are correlational. The present study will test these two hypotheses directly by manipulating the parafoveal load and parafoveal preview in single experiment, in order to distinguish the two hypotheses in terms of the accounts for the RAN deficits of children with dyslexia. The modified RAN paradigm proposed by Silva et al. (2016) was used. The stimulus set was composed of 28 RAN matrices containing 5 familiar characters. All letters occurred equally and they were pseudo randomly ordered such that there were no immediate repetitions. The spatial arrangement of the traditional 5×10-item matrix was modified, such that character items were irregularly spaced within each row. Half the items stood in the right parafoveal span of the preceding item and thus they could be parafoveally previewed (P). The other half stood beyond the parafoveal span of N-1 (non-previewed, nP). Half the items were followed by parafoveal inputs at the right (had parafoveal load, L) and the other half were not (no parafoveal load, nL). There were, thus, there were four experiment conditions (PL, PnL, nPL, nPnL). Three groups of children were selected as participants: children with developmental dyslexia (DD), chronological age-matched (CA) and reading level-matched (RL). They were instructed to name the characters presented on the screen as soon as possible while their eye movements were recorded by Eye Trackers (Eyelink 1000). The parafoveal preview benefit was computed with the formula (non-previewed—previewed)/previewed, indicating the proportion of the fixation time for previewed items (PL+PnL) that was additionally required when looking at items with no preview (nPL+nPnL). Similarly, the parafoveal load cost was computed as the proportion of the fixation time for non-loaded items (PnL+nPnL) that was additionally required when there was a parafoveal load (PL+nPL). The formula was (loaded-non-loaded)/non-loaded. All three groups of children showed significant parafoveal preview benefit for the measure of gaze duration (all ts > 4.43, all ps < 0.001), but the amount of preview benefit across three groups did not reach significance (F(2, 42) = 2.37, p > 0.05). The significant effect on the measure of total fixation time was only observed for age-matched children (t(14) = 4.70, p < 0.001) . For the amount of parafoveal load, there were no significant differences across three groups of children for measures of gaze duration and total fixation time (both Fs < 2.24, both ps > 0.05). These results above indicated that Chinese dyslexia children showed less parafoveal preview benefit than age-matched children, but parafoveal loads were not significant between the two groups. The present study yields robust evidence, supporting the first account of RAN deficit, that the parafoveal dysfunction of Chinese dyslexia children is due to their reduced parafoveal preview benefits.