OBJECTIVE: To identify items in the existing disease-specific St. George's Respiratory Questionnaire (SGRQ) that might be suitable for use in a COPD specific utility instrument. METHOD: The SGRQ has three domains. We planned to use three items per domain (mild, moderate and severe) for the new instrument. Using data from 893 patients we first reduced the original 50 SGRQ items to the 40 "best" items using classical test methods and Rasch modelling (RUMM 2010 software). The Person Separation Index for these 40 items was 0.9 (an "excellent" fit to a unidimensional model). We then used Rasch location maps to identify suitable items. We examined those items that covered 95% of the patients, then divided the population into tertiles according to their person location value: mild (mean location -0.97 logits, SD=0.35), moderate (mean location 0.09 logits, SD=0.39), severe (mean location 1.45 logits, SD=0.56). For each level of patient severity we chose one item per SGRQ domain using criteria based upon quality of fit of the item to the unidimensional model of all 40 items . RESULTS: We were able to identify one suitable item per domain at each severity level. The locations of the nine items ranged from -1.16 logits to 1.47 logits. The mean item location for the three mild items was -0.55 logits, moderate items 0.18 logits, severe items 1.16 logits. CONCLUSION: We have now identified nine items from three domains of health in COPD. Each item has a clearly defined level of severity. This approach should ensure that the utility instrument, when fully developed, has good discriminative properties and may also have good evaluative properties.