The paper presents a study in which participants learned computer literacy by having a spoken conversation with AutoTutor, an intelligent tutoring system with conversational dialogue. Thirty students completed a multiple-choice pre-test, a 35-minute training session, and a multiple-choice post-test. After completing the post-test, students reviewed their tutorial interaction and judged what emotions they experienced on the basis of the dialogue history and their facial expressions. Our results revealed that many measures of performance were impervious to poor or modest speech recognition accuracy, which is compatible with a soft constraint-based model. Speech recognition errors had a very subtle impact on learning as well as participants’ emotions and attitudes.