Background Health care workers experience high stress. Accessible, affordable, and effective approaches to reducing stress are lacking. In-person mindfulness-based interventions can reduce health care worker stress but are not widely available or accessible to busy health care workers. Unguided, digital, mindfulness-based self-help (MBSH) interventions show promise and can be flexibly engaged with. However, their effectiveness in reducing health care worker stress has not yet been explored in a definitive trial. Objective This study aimed to investigate the effectiveness of an unguided digital MBSH app (Headspace) in reducing health care worker stress. Methods This was a definitive superiority randomized controlled trial with 2182 National Health Service staff in England recruited on the web and allocated in a 1:1 ratio to fully automated Headspace (n=1095, 50.18%) or active control (Moodzone; n=1087, 49.82%) for 4.5 months. Outcomes were subscales of the Depression, Anxiety, and Stress (primary outcome) Scale short form; Short Warwick Edinburgh Mental Well-being Scale; Maslach Burnout Inventory; 15-item Five-Facet Mindfulness Questionnaire minus Observe items; Self-Compassion Scale–Short Form; Compassionate Love Scale; Penn State Worry Questionnaire; Brooding subscale of the Ruminative Response Scale; and sickness absence. Results Intention-to-treat analyses found that Headspace led to greater reductions in stress over time than Moodzone (b=–0.31, 95% CI –0.47 to –0.14; P<.001), with small effects. Small effects of Headspace versus Moodzone were found for depression (b=–0.24, 95% CI –0.40 to –0.08; P=.003), anxiety (b=–0.19, 95% CI –0.32 to –0.06; P=.004), well-being (b=0.14, 95% CI 0.05-0.23; P=.002), mindfulness (b=0.22, 95% CI 0.09-0.34; P=.001), self-compassion (b=0.48, 95% CI 0.33-0.64; P<.001), compassion for others (b=0.02, 95% CI 0.00-0.04; P=.04), and worry (b=–0.30, 95% CI –0.51 to –0.09; P=.005) but not for burnout (b=–0.19, –0.04, and 0.13, all 95% CIs >0; P=.65, .67, and .35), ruminative brooding (b=–0.06, 95% CI –0.12 to 0.00; P=.06), or sickness absence (γ=0.09, 95% CI –0.18 to 0.34). Per-protocol effects of Headspace (454/1095, 41.46%) versus Moodzone (283/1087, 26.03%) over time were found for stress, self-compassion, and compassion for others but not for the other outcomes. Engagement (practice days per week) and improvements in self-compassion during the initial 1.5-month intervention period mediated pre- to postintervention improvements in stress. Improvements in mindfulness, rumination, and worry did not mediate pre- to postintervention improvements in stress. No serious adverse events were reported. Conclusions An unguided digital MBSH intervention (Headspace) can reduce health care workers’ stress. Effect sizes were small but could have population-level benefits. Unguided digital MBSH interventions can be part of the solution to reducing health care worker stress alongside potentially costlier but potentially more effective in-person mindfulness-based interventions, nonmindfulness courses, and organizational-level interventions. Trial Registration International Standard Randomised Controlled Trial Number ISRCTN15424185; https://tinyurl.com/rv9en5kc