Affective (positive or negative) information of emotion word is able to accelerate the word’s processing compared with the neutral word. Several researches revealed that the emotional valence effect occurs under either conscious or subliminal conditions. However, if the emotional valence effect is an automatic procedure or an interactive outcome between task paradigm and stimulus property is unclear now. The present study used semantic and phonological tasks to explore the influence of frequency to emotional valence effect when reading Chinese characters. After normalization, two experiments applied the same stimuli sets of Chinese characters which had been strictly matched among their frequency, strokes number, radical number, concrete level and emotional arousal level in all 6 conditions. Experiment 1 used semantic category decision task, 2 (high frequency vs. low frequency) × 3 (negative/neutral/positive) within subjects design. Participants were 30 college students, but 1 subjects was excluded because his total error rate was too high (>10%). They were asked to decide whether the target Chinese character was living being (e.g. tiger, horse, grass) or not by pressing keyboard button. Different from semantic processing, Experiment 2 applied rhyming decision task. Another 30 college students participated in this task. They made responses according to the target characters’ pronunciation. If the syllable contained /ong/ (e.g. /song4/, /qiong2/, /nong2/) then press “yes” button, otherwise press “no”. Besides that, two experiments shared the same design, stimuli and program. Computer automatically recorded subject’s reaction time and error rate by E-prime1.1 software. The MANOVA results showed that when participants were asked to perform the semantic task the negative valence accelerated the characters' semantic access in both high and low frequency conditions (ps<.05). There was not significant interaction between frequency and emotional valence (p>.05). At the same time, the positive valence speeded up the low-frequency characters only (p<.05), but not on the high-frequency ones (p>.05). The influence of positive valence was weaker than negative valence. However, there was significant interaction between emotional valence and frequency in the phonological task (p<.001). Simple effect exam showed that emotional valence mainly impacted on the low-frequency characters' phonological access (p<.001), and didn’t affect on the high-frequency ones (p>.05). In the low-frequency condition, the RTs of negative (p<.01) and positive (p<.001) characters were shorter than neutral ones. The integration of two experiments supports the biased competition model. In summary, the results suggest that the emotional information modulate the semantic and phonological access during reading characters. Moreover, the modulation effect is influenced by the task-oriented and familiarity of material. Emotional valence can accelerate the low-frequency characters’ phonological and semantic access, but not the high-frequency characters’ phonological access. However, in semantic task, even the high-frequency characters’ processing can be affected by negative valence.