Many universities conduct student evaluations. Their purpose is to encourage improvement in teaching. However, the evaluations are merely subjective assessments by students, meaning that instructors cannot necessarily easily relate evaluations to areas for improvement in teaching. To address this issue, we suggest a teaching behavior estimation model that can estimate teaching behaviors from student evaluations of each lesson. In previous research, we built a model on the assumption that teaching behaviors are not correlated with other behaviors and that student evaluation items are uncorrelated to other evaluation items. However, this assumption could not be verified. Our research suggests a new teaching behavior estimation model that represents the correlation between factors of teaching and factors of student evaluations. To analyze this, we conducted canonical correlation between two kinds of factors and obtained correlations. This result shows that it is possible to construct a teaching behavior estimation model based on factors of teaching behavior and factors of student evaluations.