Purpose: This paper aims to investigate the relationship between student emotions, professors' performance and course ratings and difficulty. Design/methodology/approach: Natural language processing models are used to extract six basic emotions and several categories of professors' harmful performance from nearly one million student reviews randomly selected from the website ratemyprofessors.com. These features are used in regression analysis to analyse their relationship with numerical ratings of course quality and course difficulty. Findings: Negative emotions and bad performance by professors are detected more often for low-rated courses and courses perceived as more difficult by students. Positive emotions are seen for highly rated and less challenging courses. Practical implications: This paper shows that natural language processing tools can be used to enhance and strengthen the quality assurance processes at universities. The proposed methods can improve the often-contested student evaluation of teaching practices, help students make better and more informed choices about their courses and assist instructors to better tailor their teaching approaches and create a more positive learning environment for their students. Originality/value: This paper presents a novel analysis of how student emotions and poor performance by professors, derived automatically from teacher evaluations by students, affect course ratings. Results also lead to a novel hypothesis that the student-course emotional match or student tolerance of bad behaviour by professors can affect the performance of students and their chances of completing their degree.