Students' evaluations of teaching are increasingly used by universities to evaluate teaching performance. However, these evaluations are controversial mainly due to the fact that students value various aspects of excellent teaching differently. Therefore, in this paper we propose a new approach to students' evaluations of university teaching based on data from conjoint analysis. Conjoint analysis is a multivariate technique used to analyze the structure of individuals' preference. In particular, our approach accounts for different importance students attach to various aspects of teaching. Moreover, it accounts explicitly for heterogeneity arising from students' preferences, and incorporates it to form comprehensive teaching evaluation score. We have conducted survey and confirmed applicability and efficiency of the proposed approach.