It is well-documented that ratings of medical care received personally (personal referent) yield more favorable responses than ratings of care received by people in general (general referent). Hence general items are useful in achieving greater variation in responses to satisfaction surveys. However, the validity of general items relative to personal items is being debated currently. It has been hypothesized that bias due to socially desirable response set (SDRS) would be greatest for items with a personal referent. To test this hypothesis, the authors compared both kinds of satisfaction ratings for adults (N = 3,918) who scored high and low on SDRS during Rand's Health Insurance Experiment. Across sites and years of the experiment, the rating item with a personal referent was consistently biased upward for those manifesting SDRS. The rating item with a general referent was not. Further, the correlation between SDRS and the difference between ratings on the personal and general referent items was statistically significant, suggesting that more favorable ratings of medical care received personally compared with ratings of care received by people in general are in part due to SDRS bias. Results are discussed in terms of implications for constructing a valid satisfaction survey.