OBJECTIVES: EQ-5D defines health in terms of 5 dimensions each divided into 3 levels, forming a classification of 245 states. The first level within each dimension corresponds to “no problem” and it has been suggested that the classification lacks scope for sensitivity. A 5-level descriptive classification offers advantages with smaller step sizes than is presently the case, leading to reduced ceiling effects. This paper reports on a study that compares self-rated health status in a patient survey using a modified 5-level descriptive classification alongside the conventional 3-level EQ-5D. METHODS: EQ-5D was modified by inserting an intermediate level between the existing levels 1&2 and 2&3. No text labels were provided for these levels 2 and 4 of the modified version. A questionnaire containing both 3-level and 5-level response systems was constructed. Order of presentation was varied with 50% presenting the standard EQ-5D format first; in the remainder the 5-level version was presented first. Questionnaires were mailed out to over 2000 individuals selected from the national electoral register. RESULTS: Data from 950 respondents were available for analysis (n=478 for 3-5 version and 472 for the 5-3 version respectively), of which 64% of respondents were indifferent when asked which version they preferred. Of the remainder 65% preferred the 5-level version. Response errors were defined as a difference in response of more than 1 level between the 2 versions and these were generally less than 1%. There was a smaller ceiling effect and only 39% of respondents reported no problems on the 5-level version. The corresponding rate for the 3-level version was 50%. CONCLUSIONS: A 5-level response system for EQ-5D is feasible. The revised version produced a wider distribution of reported problems with less compacting towards level 1. Backwards compatability ensures that existing valuation sets can be used with both 3 and 5-level systems.