Heart rate variability analysis can play an important role in emotion recognition. In order to establish an accurate model between ECG and emotion categories, a feature selection method based on maximum information coefficient (MIC) is proposed. In this paper, the Aubt database and the design of emotional induction experiments are used for research. First, 40 features based on heart rate variability in time domain, frequency domain, nonlinear and time-frequency domain were extracted, then emotion modelingwas conducted based on the MIC method combined with support vector machine, random forest and K nearest neighbor algorithm. The results show that based on the MIC feature selection method,the classification accuracy of the Aubt database for arousal, valence, and four emotions is 90%, 89%, and 84%, respectively. And further choose Pearson correlation coefficient, ANOVA feature selection method to compare with MIC. In the induced experimental data,the correct classification rate ofmultiple one-to-one emotion recognition is higher than 75%. It shows that the MIC feature selection method can significantly improve the classification accuracy, which is of great significance for emotion recognition based on ECG signals.