The aim of this study was to examine fit of the original 24-item Roland-Morris Disability Questionnaire and three short-form versions to a Rasch model with particular attention to targeting of item difficulty and to differential item functioning.Cross-sectional survey of 140 people with low back pain seeking physiotherapy treatment. Data were analysed using a dichotomous Rasch model.All versions showed adequate overall data fit to the Rasch model, with few misfitting items. Person separation was around 0.85 for all versions. Item 5 (use a handrail to get upstairs) showed differential item functioning by age. Targeting of persons of high ability was poor and short-form versions also had poor targeting of persons of low ability. Items of similar difficulty clustered in the centre of the logit scale.Although the Roland data fit the Rasch model, there were insufficient items of higher difficulty to sufficiently evaluate disability in persons with mild disability. Short-form versions also lacked items of lower difficulty.