Purpose: This study aimed to explore how autonomy support mediates the association between exercise knowledge and exercise self-efficacy in patients with severe mental illness. Methods: A total of 181 individuals were selected through convenience sampling. They completed self-administered surveys measuring exercise knowledge, autonomy support, and exercise self-efficacy. The data analysis was conducted using descriptive statistics, t-tests, one-way analysis of variance, Pearson’s correlation coefficients, and multiple linear regression. To determine the statistical significance of the mediating effect, a bootstrapping test was also conducted. Results: This research revealed that autonomy support significantly predicted exercise self-efficacy and partially mediated the association between exercise knowledge and exercise self-efficacy (lower limit confidence interval [LLCI]: 2.34; upper limit confidence interval [ULCI]: 5.97). The model incorporating exercise knowledge and autonomy support explained 42.0% of the variance in exercise self-efficacy. Conclusion Collaborative endeavors involving mental health practitioners and exercise experts demonstrate notable efficacy in enhancing exercise self-efficacy of individuals with mental illness.