Research on second-language (L2) speech acquisition suggests that audio-visual aids could be effective in helping learners acquire difficult L2 speech sounds (Li, 2016a). However, most previous studies have been restricted to laboratory settings rather than the classroom environment. The present study, therefore, was designed to fill this knowledge gap by analysing the effectiveness of audio-visual aids, particularly articulatory gestures, in teaching L2 speech sounds in actual classrooms. The participants were students from two classes of non-English majors who had severe difficulties with the differentiation of /[theta]/--/s/ and /[voiced dental fricative]/--/z/. 'Read-aloud' tasks were employed for pronunciation tests. The baseline data of the students' pronunciation of the two contrasts was collected with a pre-test, and the intuitive--imitative approach was adopted for teaching. Specifically, the students were exposed to seven audio- or audio-visually recorded poems (one poem per week over the course of seven weeks) containing the target contrasts. The students in Class 1 were taught with the audio-recordings without images of the speaker's face being displayed; in contrast, students in Class 2 were taught using audio-visual recordings, which allowed them to observe the speakers' articulatory gestures of /[theta]/--/s/ and /[voiced dental fricative]/--/z/. To detect the teaching effect, a post-test was carried out after the teaching programme was completed. A delayed post-test was conducted one month after the post-test. Comparisons with the respective pre-test results indicated that students in Class 2 had a significant improvement in the pronunciation of the target contrasts in the post-test, whereas those in Class 1 did not. In the delayed post-test, neither class showed any significant difference in the pronunciation performance in comparison with the post-test. The findings, therefore, confirmed the effectiveness of exposure to audio-visual aids in teaching L2 pronunciation.