Language comprehension not only refers to lexical access of an individual word, but also involves semantic integration processing, which builds up a coherent mental representation of isolated words. Previous studies using eye tracking or ERPs methods have investigated the time course of semantic integration, but their results remained controversial. Evidence from eye tracking showed that the earliest semantic integration occurred before 250ms. But in most ERPs studies, semantic integration is reflected by classic N400 effect between 300 and 500 ms, which is later than 250ms. A potential account of the controversy is that compared to late components, the early ones are typically focal and short in duration. Therefore, they are more likely to be affected by large stimulus variance on relevant linguistic parameters. Though important physical and psycholinguistic parameters were matched in most previous studies, attention paid to the range of relevant parameters of stimuli was not enough. By limiting the frequency variance of critical words to a very small range, the present ERPs experiment was conducted to examine whether early ERPs components were linked to semantic integration during Chinese reading comprehension. A hundred Chinese sentences were constructed for the present study. Each critical word at the end of the sentence was counterbalanced between semantically congruent and incongruent conditions. Especially, the frequency variance of critical words was minimized as small as possible. Then sentences were presented word by word at the centre of the screen, and each word has a duration of 400 ms with an inter-stimuli interval (ISI) of 200 ms. After each sentence, participants were asked to decide whether the sentence was plausible or not. We recorded 18 participants' ERPs with 32 electrodes of Neuroscan system. Epochs of interest were time-locked to the onset of critical words with a 200 ms pre-onset baseline window and an 800 ms post-onset window. P2 and N400 were computed in the time windows of 150~300 ms and 300~500 ms. A more positive P2 and a less negative N400 component were found when the sentence context was congruent than they were in incongruent condition, though P2 effect was smaller than N400 effect obviously. In conclusion, the present study supports that by minimizing the relevant variance of critical words to a small range, semantic integration can be detected before 250 ms in ERPs studies. Though P2 effect reflecting semantic integration is transient and influenced by lexical information easily. In contrast, semantic violation condition evokes stronger N400 component, which is enduring and stable. This finding may help to resolve the conflict between eye tracking and ERPs results.