Background: Teachers' self-reported stress is related to the quality of teacher-student interactions and students' learning outcomes. However, it is unclear if teachers' physiological stress is related to child-centred teaching practices in the classroom and whether teaching practices mediate the link between teachers' stress and students' learning outcomes. Aims: We studied the effect of teachers' physiological stress and self-reported stress on their teaching practices and thereby on students' learning outcomes in math. Sample: A total of 53 classroom teachers and 866 Grade 1 students participated in the study. Methods: Salivary cortisol in the middle of the school day and cortisol slope from morning peak to evening were used as indicators of teachers' physiological stress, in addition to self-reported teaching-related stress. Teaching practices were observed with the ECCOM instrument. Students' math skills controlled for gender and previous skills were used as a measure of learning outcomes. Data were analysed with a two-level SEM. Results: Teachers' physiological stress did not have an effect on teaching practices or students' math skills. Teachers reporting less stress used relatively more child-centred teaching practices compared with teacher-directed ones. These practices had a marginal effect on classroom-level differences in the gain of students' math skills in Grade 1. There was neither a direct nor indirect effect from teachers' stress on students' math skills. Altogether, our model explained 77% of classroom-level variance in math skills. Conclusions: Teachers' self-reported stress has an effect on their teaching practices, which, in turn, have a marginal effect on students' learning outcomes.