Previous research has suggested that higher levels of teachers' self-efficacy (TSE) tend to be positively related to positive teachers' emotions (e.g., joy, pride) and negatively to negative teachers" emotions (e.g., anger, anxiety). However, these studies predominately relied on cross-sectional design and therefore were unable to test the reciprocal relations between the two constructs. Based on the propositions of social-cognitive theory (Bandura, 1997), TSE may be viewed as an antecedent or as a consequence of emotions. More specifically, TSE may shape emotions since it directs teachers' attentional, appraisal, and regulatory processes, while emotions may shape TSE since they act as a source of information about teachers' performance in a given task (i.e., emotions can serve as a filter that determines which efficacy information is seen as salient and how it is interpreted). To test these assumptions, an initial sample of 3010 Croatian teachers (82% female) participated in a longitudinal study based on a full panel design with three measurement points and time lags of approximately 6 months. Teachers taught at different educational levels (i.e., elementary, middle, and secondary schools) and had on average 15.30 years ( = 10.50) of teaching experience. They completed self-report measures that assessed their self-efficacy beliefs and six discrete emotions experienced in relation to teaching and students - joy, pride, love, anger, hopelessness, and exhaustion. An autoregressive cross-lagged analysis showed that teachers' emotions and TSE are indeed related to each other. However, the direction of this association is not bidirectional as was suggested by theoretical assumptions; instead, it is asymmetrical - higher levels of TSE beliefs predicted higher levels of positive emotions of joy and pride, while higher levels of teachers' negative emotions of anger, exhaustion, and hopelessness predicted lower levels of teachers' self-efficacy beliefs.