Mindfulness training (MT) is considered appropriate for school teachers and enhances well-being. Most research has investigated the efficacy of instructor-led MT. However, little is known about the benefits of using self-taught formats, nor what the key mechanisms of change are that contribute to enhanced teacher well-being. This study compared instructor-led and self-taught MT based on a book (Williams & Penman, 2011) in a sample of secondary school teachers. We assessed expectancy, the degree to which participants believed the intervention was effective, their program engagement, well-being and psychological distress, and evaluated whether mindfulness and self-compassion skills acted as mediators of outcomes. In total, 206 teachers from 43 schools were randomized by school to an instructor-led or self-taught course-77% female, mean age 39 years (SD = 9.0). Both MT formats showed similar rates of participant expectancy and engagement, but the instructor-led arm was perceived as more credible. Using linear mixed-effects models, we found the self-taught arm showed significant pre-post improvements in self-compassion and well-being, while the instructor-led arm showed such improvements in mindfulness, self-compassion, well-being, perceived stress, anxiety, depression, and burnout. Changes over time significantly differed between the groups in all these outcomes, favoring the instructor-led arm. The instructor-led arm, compared with the self-taught, indirectly improved teacher outcomes by enhancing mindfulness and self-compassion as mediating factors. Mindfulness practice frequency had indirect effects on teacher outcomes through mindfulness in both self-taught and instructor-led arms. Our results suggest both formats are considered reasonable, but the instructor-led is more effective than the self-taught. Trial registration: ISRCTN18013311.