Aim: To examine resilience in Science, Technology, Engineering, and Mathematics (STEM) learning within an ecological model, identifying the psychological processes associated with resilient, and non-resilient learning to develop a framework for promoting STEM resilience. Sample and method: From a sample of secondary-school students (n = 4,936), 1,577 students who found their STEM lesson difficult were identified. Students were assessed on three resilience capabilities and asked to write a commentary on how they responded to the lesson. Results: Factor analysis revealed that resilience in STEM learning could be positioned within the ecological systems model, with students' resilience being comprised of three capabilities; the ability to quickly and easily recover (Recovery), remain focussed on goals (Ecological), and naturally adjust (Adaptive capacity). Using a linguistic analysis programme, we identified the prevalence of words within the student commentaries which related to seven psychological processes. Greater ability to recover was negatively related to negative emotional processes. To increase the specificity of this relationship, we identified high and low resilient students and compared their commentaries. Low resilient students used significantly more anger words. Qualitative analysis revealed interpersonal sources of anger (anger at teacher due to lack of support) and intrapersonal sources of anger (including rumination, expression and control, and seeking distraction). Conclusions: Anger is a key process that distinguishes students who struggle to recover from a difficult STEM lesson. An ecological systems model may prove useful for understanding STEM resilience and developing intervention pathways. Implications for teacher education include the importance of students' perceptions of teacher support.