Formative assessment has been identified as a promising intervention to support students’ learning. How to successfully implement this means of assessment, however, is still an open issue. This study contributes to the implementation of formative assessment by analyzing the impact of a training measure on teachers’ formative feedback practice, with a special focus on mediating and moderating variables. Research questions are as follows: (1) Is there an indirect training effect on teachers’ instructional feedback practice via (a) teachers’ declarative feedback knowledge and (b) the ability to generate feedback in a test situation? (2) Is this indirect effect moderated by teachers’ self-efficacy? A total of 67 secondary education mathematics teachers participated in the study, taking part in professional development either on formative assessment and feedback (PD-FA) or on mathematical modelling and problem solving (PD-PM). Training was provided in two sessions (T1 and T2; each lasting 3 days) with 10 weeks in between T1 and T2. Teachers’ self-efficacy regarding feedback was measured before T1 with a questionnaire. Declarative feedback knowledge and the ability to apply this knowledge were tested after T2. Teachers’ instructional feedback practice was assessed with a student questionnaire (before T1 and 4–6 weeks after T2). Path analyses show that (1) there is no indirect training effect (PD-FA vs. PD-PM) on the development of teachers’ feedback practices in mathematics instruction; but an indirect effect on the ability to generate feedback in a test situation via teachers’ declarative feedback knowledge. Teachers participating in PD-FA show a higher level of declarative feedback knowledge than teachers in the PD-PM condition. Declarative feedback knowledge in turn is positively related to the ability to generate feedback in a test situation. (2) This indirect effect is moderated by teachers’ self-efficacy. Teachers with a high level of self-efficacy are better able to use their knowledge to generate feedback in a test situation than teachers with a low level of self-efficacy.