The study manily concerned the top-down modulation on automatic processing, which could modify the classical automatic processing theory and give suggestions to implicit study methods to some certain. Top-down modulation is the function of our brain to modulate and select task-related information and allocate our attention on it, so we can accelerate the information processing. Automatic processing is defined as a process that needs little attention and cognitive resources comparing with the controlled processing. Recent findings questioned classical automatic processing theory by telling that maybe the automatic processing is also modulated by top-down factors including intention, task-set, temporal attention, spatial attention, expectation and etc. We improved the previous studies by modifying the paradigm and using Chinese stimuli. The study used masked priming paradigm to investigate if the top-down task-set factor could modulate the automatic semantic processing by ERPs. Subjects were asked to perform categorization task or structure decision task of the former stimuli (primes) and lexical decision task of the latter ones (targets). So, there are two kinds of task-sets: semantic and perceptual task-set. The stimuli were Chinese two-character words and pseudo-words citing from other articles. And the masks were a series of “#”. 14 undergraduate students were chose as subjects (6 males, 8 females). The mean age (±SD) is 21.9±(1.6) . The mean amplitude and peak latency of N400 which epoched between 300-450ms at F3、FZ、F4、C3、CZ、C4、CP3、CPZ、CP4、P3、PZ、P4、PO3、POZ、PO4 electrodes and the RT (response time) were used as indexes. The behavioral data and the electrophysiological data were acquired simultaneously by software Stim 2 and Scan 4.2. We used Repeated-ANOVA. The experimental design was 2 (task factor: semantic task & perceptual task) × 2 (priming conditions: prime-target semantic related & unrelated). And the electrode factor were introduced when statistical analysis were proceeded. The main behavioral result was that the main effect of task factor was significant, F（1，13）=7.703，p< .05, and the reciprocal effect between task factor and priming conditions was significant, F（1，13）=6.354, p< .05. Under the condition of semantic task-set, the RT of prime-target semantic related （838±130 ms）vs. unrelated（814±126 ms）was significantly different, F（1，13）=8.631, p< .05. The main result of electrophysiological data was : the reciprocal effect between task factor and priming conditions was significant, F（1，13）=4.819，p< .05. By further analysis, we found that under the condition of semantic task-set, the amplitude under the prime-target semantic related vs. unrelated conditions were significantly different, F（1，13）=7.167, p< .05, Mrelated=4.230?v, SD= .855?v, Munrelated=3.054?v, SD= .817?v. The conclusion is automatic semantic activation can be modulated by top-down task-set factor.