The Roland-Morris Disability Questionnaire (RM24) contains 24 items, patients completing the questionnaire by ticking the items that apply to them. It is likely that requiring a patient to respond in this manner may lose information about the patient's disability. Developing a similar questionnaire with a multi-level response option would address this. Therefore, the aims of this study were to develop a modified version of the RM24 (multi-level RM24) and to evaluate its measurement properties. In a series of three studies involving 667 patients with low back pain the descriptive statistical properties of the multi-level RM24 and the psychometric properties of the original and multi-level RM24 were evaluated. The multi-level RM24 was shown to have good internal consistency, reliability and cross-sectional construct validity, and to minimize the ceiling effect found in the RM24. The longitudinal construct validity for the multi-level RM24 was also greater than for the RM24. In conclusion, replacing the yes/no response option with a multi-level response option improves the measurement properties of the RM24.