It has been suggested that the relatively poor effectiveness of treatments for anorexia nervosa is due to a poor conceptualisation of the disorder. One hypothesis is that current models are mistakenly targeting superficial, instead of deeper level, cognitions and cognitive processes. A schema-based cognitive-behavioural model of eating disorder pathology suggests that the process of schema compensation is key to restrictive pathology-when there is the threat of experiencing negative affect, compensatory schemas are activated, reducing that affect. The current experimental study aimed to provide support for such a process. Eating-disordered and control women completed a computer-based task, measuring the compensation process in terms of speed and accuracy in response to subliminal threat cues. The results did not fully support the hypothesis, suggesting that the model and methodology need some amendment. Improvements to the methodology are discussed.