A consistent finding in health services research is the report of uneven uptake of research findings. Implementation trials have a variable record of success in effectively influencing clinicians' behaviour. A more systematic approach may be to conduct Intervention Modelling Experiments before service-level trials, examining intervention effects on 'interim endpoints' representing clinical behaviour, derived from empirically supported psychological theories. The objectives were to: (1) Design Intervention Modelling Experiments by backward engineering a 'real-world' randomised controlled trial (NEXUS); (2) examine the applicability of psychological theories to clinical decision-making; (3) explore whether psychological theories can illuminate how interventions achieve their effects. A 2 x 2 factorial randomised controlled trial was designed with pre- and post-intervention data collection by postal questionnaire surveys. The first survey was used to generate feedback data and the interventions were delivered in the second survey. General medical practitioners (GPs) in England and Scotland participated. First survey respondents were randomised twice to receive or not audit and feedback and educational reminder messages. The main outcome measures included behavioural intention (general plan to refer for lumbar X-rays) and simulated behaviour (specific, scenario-based, decisions to refer for lumbar X-ray). Predictors were attitude, subjective norm, perceived behavioural control (theory of planned behaviour), self-efficacy (social cognitive theory) and decision difficulty. Both interventions significantly influenced simulated behaviour, but neither influenced behavioural intention. There were no interaction effects. All theoretically derived cognitions significantly predicted simulated behaviour. Only subjective norm was not predictive of behavioural intention. The effect of audit and feedback on simulated behaviour was mediated through perceived behavioural control. The results of this study suggest that Intervention Modelling Experiments, using psychological models to help isolate mediators of clinical decision-making, may be a means of developing more potent interventions, and selecting implementation interventions with a greater likelihood of success in a service-level randomised controlled trial.