To improve interventions that aim to promote return to work (RTW) of workers with common mental disorders (CMD), insight into modifiable predictors of RTW is needed. This study tested the predictive value of self-efficacy change for RTW in addition to preintervention levels of self-efficacy. RTW self-efficacy was measured 5 times within 9 months among 168 clients of a mental healthcare organisation who were on sick leave due to CMD. Self-efficacy parameters were modelled with multilevel analyses and added as predictors into a Cox regression analysis. Results showed that both high baseline self-efficacy and self-efficacy increase until full RTW were predictive of a shorter duration until full RTW. Both self-efficacy parameters remained significant predictors of RTW when controlled for several relevant covariates and within subgroups of employees with either high or low preintervention self-efficacy levels. This is the first study that demonstrated the prognostic value of self-efficacy change, over and above the influence of psychological symptoms, for RTW among employees with CMD. By showing that RTW self-efficacy increase predicted a shorter duration until full RTW, this study points to the relevance of enhancing RTW self-efficacy in occupational or mental health interventions for employees with CMD. Efforts to improve self-efficacy appear valuable both for people with relatively low and high baseline self-efficacy.