Research on earnings inequality mostly relies on survey data, but these data may not be accurate. Survey data on earnings might be biased as research indicates that some respondents are likely to avoid reporting their gross earnings and others are likely to misreport them. In addition, the mode of data collection might affect responses to sensitive questions such as those on earnings. Given these three possibilities for bias, researchers’ conclusions on the degree of earnings inequality might be systematically biased as well. By comparing survey and linked administrative data, we looked for the nonreporting and misreporting biases suggested by the literature, investigated the presence of an interviewer as another source of non- and misreporting, and compared how nonreporting, misreporting, and the mode of data collection affected conclusions on earnings inequality. The analyses drew on a German employee survey and linked administrative data from the Federal Employment Agency. Using the administrative data as a benchmark, we found that respondents at the lower and upper end of the earnings distribution were more likely to not report and to misreport their earnings. Interviewer presence led to higher nonreporting but had no effect on misreporting. All these processes and especially nonreporting and interviewer presence led to an underestimation of earnings inequality based on survey data. We relate the relevance of these results to research on inequality and survey methodology and conclude that linking survey data to administrative records could be an avenue for safeguarding conclusions on earnings inequality.