Objective To explore the effects of embodied emotion training and the emotion training of social cognitive interaction training on emotion recognition in schizophrenic patients. Methods A total of 45 patients with schizophrenia in Xilingol League Anshen Hospital from January to April 2022 were selected for the study. The randomized numerical table method was used to divide study subjects into 15 cases in embodied emotion group， 15 cases in emotion training group and 15 cases in blank control group. The intervention was carried out for a period of five weeks， three times per week. Among them, the embodied emotion group received embodied emotion training, the emotion training group received emotion training, and the blank control group did not receive any intervention. Before and after treatment， patients' emotion recognition skills were assessed using the Emotion Recognition Task. Friedman ANOVA was used to compare the differences in correct emotion recognition rates on pre- and post-tests between the three groups， and a repeated-measures analysis was performed using generalized estimating equations modeling to analyze the correct emotion recognition rates of the three groups of patients. Results Friedman ANOVA showed that the median correct posttest rate of patients in emotion training group in positive mood faces （0.95） was higher than that in the pretest （0.80）， and the difference was statistically significant （Z=5.333， P＜ 0.05）. The median correct posttest rate of patients in emotion training group in negative mood faces （0.80） was higher than that in the pretest （0.55）， and the difference was statistically significant （Z=14.000， P＜ 0.05）. The median correct posttest rate of patients in blank control group in negative emotional faces （0.50） was lower than that of the pretest （0.70）， and the difference was statistically significant （Z=8.067， P＜ 0.05）. Generalized estimating equation modeling revealed a significant level of the time × group effect in the neutral and negative emotion valences （Wald χ2 =11.631， P＜ 0.05； Wald χ2 =49.740， P＜ 0.05）. Further analysis of separate effects showed higher correct rates after training than before training in both positive and negative emotion valences （both P ＜ 0.05）. Separate effect analysis by group suggested that embodied emotion training group had higher correct rate than blank control group in the neutral emotion valence （P＜0.05）. Conclusions The emotion training of social cognitive interaction training can improve emotion recognition in schizophrenia， and embodied emotion training can maintain patients' neutral emotion recognition and avoid the tendency of decline.