Purpose: The study aimed to understand the effect of instructional coaching on teachers' implementation of a science teaching improvement programme and whether it varies in schools of different socioeconomic statuses. Design/methodology/approach: The authors conducted an experimental study. A total of 59 seventh-grade classrooms from a representative sample of public schools from the city of Buenos Aires, Argentina, were provided with research-based science educative curriculum materials (ECM) as resources to improve their teaching. A randomly selected treatment group received additional instructional coaching. Coaches met one-on-one with teachers on a weekly basis, providing pedagogical support to enact the ECM. After a 12-week intervention, the authors analyzed science teaching practices as evidenced in students' notebooks. The authors used a fidelity framework to understand the programme's implementation (with and without coaching), considering its adherence, dosage and quality, and compared how it varied across schools. Findings: While teachers in both groups used the ECM in their science lessons (i.e. with high adherence), instructional coaching almost tripled science teaching time (i.e. the dosage) but did not increase the quality of implementation (i.e. the percentage of inquiry-based science activities taught). In low socioeconomic status schools, the effect of coaching on dosage was even more intense. Originality/value: This study provides robust evidence on the impact of instructional coaching on teaching improvement programmes in science in developing countries, an under researched topic. The findings may contribute to developing targeted coaching interventions considering their effectiveness in different school contexts.