This article makes a systematic investigation on emotional forecasting accuracy of teacher on the basis of the typical education situations through three studies. The three studies were conducted in the school to the scene, including two stages of teachers' predicted emotion measurement and students' emotional experience measurement, which take the design of teacher-student paired study. We invite 21 teachers to predict the tense feeling of the paired 8 or 10 students in the 1st study; and 16 teachers forecast two paired students’ pleasant and shame mood when faced with different parenting style in the 2nd one. The third group has 18 teachers and the corresponding number students, and the teachers forecast the students’ positive high-arousal and positive low-arousal emotions when they are subject to positive expectations. The three studies have multi-samples, are cross-situational, and take different emotion measurement methods, which show that: in a typical situation of education, teachers fail to accurately predict student emotions, and there appear apparent bias in their prediction; what’s more, compared with the basic emotion, teachers predict more inaccurately to students’ complex emotions ; compared with positive low-arousal emotions, teacher predict more inaccurately to students’ positive high-arousal emotions. Despite of inaccuracies of teachers' emotional forecasting to students’ emotion, there is not obvious overestimation or underestimation of trend. The research is close to current educational practice, has good ecological validity, and tries to explore the emotion measurement method, accuracy of statistical indicators, but there are also certain limitations, i.e. a small amount of samples.