OBJECTIVES: The short form 36 (SF-36) health questionnaire may not be appropriate for population surveys assessing health gain because of the low responsiveness (sensitivity to change) of domains on the measure. An hypothesised health gain of respondents in social class V to that of those in social class I indicated only marginal improvement in self reported health. Subgroup analysis, however, showed that the SF-36 would indicate dramatic changes if the health of social class V could be improved to that of social class I. DESIGN: Postal survey using a questionnaire booklet containing the SF-36 and a number of other items concerned with lifestyles and illness. A letter outlining the purpose of the study was included. SETTING: The sample was drawn from family health services authority (FHSA) computerised registers for Berkshire, Buckinghamshire, Northamptonshire, and Oxfordshire. SAMPLE: The questionnaire was sent to 13,042 randomly selected subjects between the ages of 17-65. Altogether 9332 (72%) responded. OUTCOME MEASURES: Scores for the eight dimensions of the SF-36. STATISTICS: The sensitivity of the SF-36 was tested by hypothesising that the scores of those in the bottom quartile of the SF-36 scores in class V could be improved to the level of the scores from the bottom quartile of SF-36 scores in class I using the effect size statistic. RESULTS: SF-36 scores for the population at the 25th, 50th, and 75th centiles were provided. Those who reported worse health on each dimension of the SF-36 (ie in the lowest 25% of scores) differ dramatically between social class I and V. Large effect sizes were gained on all but one dimension of the SF-36 when the health of those in the bottom quartile of the SF-36 scores in class V were hypothesised to have improved to the level of the scores from the bottom quartile of SF-36 scores in class I. CONCLUSIONS: Analysis of SF-36 data at a population level is inappropriate; subgroup analysis is more appropriate. The data suggest that if it were possible to improve the functioning and wellbeing of those in worst health in class V to those reporting the worst health in class I the improvement would be dramatic. Furthermore, differences between the classes detected by the SF-36 are substantial and more dramatic than might previously have been imagined.