Effects of structured forgiveness training on psychological, spiritual and physiological variables were studied in healthy pre-menopausal women (N = 63, mean age 38.6 years) randomized into a four-session psychoeducational group intervention or a wait-list condition. The intervention focused on forgiving interpersonal hurts or grievances. Salivary cortisol samples, measures of anger, depression, hostility, hope, and affect, were collected to determine if forgiveness training could reduce physical and psychosocial factors often linked to health risk. While forgiveness studies have found psychosocial correlates to be related to health status, most have used cross-sectional designs and relied exclusively on brief questionnaires. This is the first study to use a randomized design to assess a possible link of with cortisol and dehydroepiandrosterone (DHEA). Results showed significant reductions (p < .007, ES = 1.17) in morning cortisol in the treatment condition, controlling for baseline level while changes in nighttime cortisol at post-testing were marginally significant (F (2,63) = 3.704, p < .059).Only morning results of DHEA approached significance (p < .15). Multiple regression showed baseline Positive Relations with Others (Ryff, 1989) accounted for reductions in post-morning cortisol (p < .016) for both conditions. Total Forgiveness increased (p < .001, ES = 0.74) for treatment participants at six-week follow-up as did Forgiveness Self-Efficacy (p < .01, ES = 0.98). However, self-efficacy failed to predict hormonal or psychological outcomes. Overall negative affect decreased significantly compared to controls. Hostile cognitions decreased (p < .001 ES=.31). Positive affect increased: Hope and Positive States of Mind increased at post treatment (p < 0.05) but not at follow-up. Baseline spiritual variables (frequency of attending services, prayer, meditation) did not predict changes in cortisol or DHEA. Frequency of prayer and meditation predicted higher levels of forgiveness at post-testing (Beta = .207, t = 2.4, p < .020), but not a follow-up, and predicted less Derogation of Others (Beta .251, t = 2.370, p < .022).