Nonresponse bias occurs when the answers of respondents are different from the presumed answers of nonrespondents. Its analysis is critical for surveys with low response rates. Usually, the analysis is limited to adjusting for information gathered in the early stages of data collection. To study nonresponse for a recent inpatient satisfaction survey, RTI conducted a telephone follow-up using a supplement of survey questions. We then implemented three approaches: comparing survey responses for sample frame attributes with significantly different response rates, comparing survey responses to the second questionnaire mailing with those for the first mailing and to the overall data, and comparing survey outcome variables between mail survey respondents and mail survey nonrespondents. The results from these three approaches provided strong evidence for the presence of nonresponse bias.