The invention provides an obtaining method of a classifier for emotion recognition. The obtaining method comprises the steps: obtaining a first sample set and a second sample set, Wherein each elementin the first sample set and the second sample set comprises four contents, namely a first characteristic value based on a heart rate variation rate, a second characteristic value based on a skin conductance signal, a third characteristic value based on the skin conductance signal and an emotion classification result; Acquiring a training process control parameter and initializing a cyclic controlparameter and a classifier; Executing a loop training process to train sub-classifiers meeting preset requirements; Cascading the sub-classifiers with a current classifier to update the current classifier; Updating the misjudgment rate and the detection rate of the current classifier; Judging whether the misjudgment rate of the current classifier is greater than the maximum misjudgment rate or not; And if not, outputting the current classifier. The negative emotion recognition method focuses on misjudged samples and samples expressing negative emotions, so that the classifier can absorb wrongteaching training, and the recognition sensitivity to the negative emotions is improved.