A combination of detailed observation of dressing behaviour and neuropsychological assessment was used to identify the cognitive barriers to independent dressing in a series of eight stroke patients. For those with right hemisphere damage, dressing was disrupted by visuospatial problems or poor sustained attention. Those with left hemisphere damage and ideomotor apraxia were unable to learn the correct procedure to compensate for hemiparesis when dressing. The value of a therapeutic approach based on these observations was assessed in a single-blind randomised multiple-baseline experiment. A baseline phase of conventional dressing therapy which takes no account of individual patterns of cognitive impairment was contrasted with an approach which was formulated for each case. A permutation test demonstrated that there was a significant treatment effect for the right hemisphere cases but there was no therapy-related improvement for those with left or bilateral damage and apraxia. Observation of a naturalistic but controlled task (dressing with a standard item of clothing) appears to allow greater insight into the impact of specific neuropsychological deficits than has been found for more complex naturalistic tasks. The benefits of this ecological approach over conventional approaches to dressing therapy need to be evaluated further in a randomised clinical trial.