Emotion plays a very important role in daily life of human beings. One of its most useful functions is to elicit full preparations for adaptive reactions. However, to develop effective behavioral strategies needs successful and effective encoding of emotional stimuli. There have been many researches on emotion processing about visual channel stimulus, such as facial expression, emotional pictures and so on. Researches on emotional speech processing are relatively few. In addition, whether processing of emotional prosody can be modulated by selective attention is still controversial. Therefore, this study aimed to explore the neural mechanism of emotional prosody processing and what roles can experimental task play on it. Two experiments were carried out using both event-related potentials (ERPs) and behavioral measures. In experiment 1, the task was emotion identification of the speech and in experiment 2, the task was gender identification of the speakers. Three kinds of emotional stimuli, ie., happy, neutral and anger prosody were used in both experiments, which were produced by two professional speakers, a male and a female news broadcaster. 23 college students and postgraduates participated in the first experiment, the other 25 participated in the second. Behavioral results showed that: Identification of emotional stimulus was faster than neutral stimulus in both experiments. And there was a negative bias of valence effect, reaction to anger stimuli was faster than to happy stimuli. ERP results showed that: (1) In 175-275ms, which is the time window of P2 component, emotional prosody processing can be modulated by experiment tasks. In the first experiment of emotion identification task, there was valence effect and negative bias, and anger emotion evokes more positive P2 component, but in the second experiment of gender identification task, no valence effect was found; (2) In the late stage of evaluation processing and response preparation, anger stimuli evoked more positive wave than neutral and happy emotion stimuli in both experiments, no significant difference was found between happy and neutral prosody, indicating that the negative stimuli had stronger stimulate correlation, thus get much more in-depth processing. These results suggested that, there were different cognitive mechanisms of emotional speech processing, compared with the neutral stimuli, emotional prosody can be processed more effectively. And to some extent, this process can be modulated by experimental tasks, especially in the second stage.