There has long existed the argument that science teachers' self-efficacy directly correlates with instruction. Regarding self-efficacy for culturally responsive instruction (CRI), a promising equitable education approach, it is posited that lack of CRI exposure in coursework contributes to low self-efficacy and poor implementation. Yet, research has illuminated instances in which self-efficacy does not relate to performance, thus leading to uncertainty about the importance of cultivating teachers' self-efficacy for CRI. This explanatory sequential mixed methods study compared longitudinal data of 19 US secondary science teachers' self-efficacy for CRI with observed practices. Participants completed a graduate-level induction programme on culturally responsive and reflective science instruction at time of study. Data were collected in the form of self-assessments, video recordings of classroom practices, and semi-structured interviews. Results showed science teachers' self-efficacy for CRI was high and remained stable throughout the programme. When comparing CRI self-efficacy to observed instruction, science teachers scored themselves higher than did researchers. This discordance was significant and did not resolve during the programme, except sociopolitical consciousness-raising instruction. Interviews were explored to determine potential sources of discordance. These included forcible disincentives and misjudgment of CRI capability as being inconsequential. Implications for culturally responsive science teacher preparation are discussed.