The income of the self-employed is often assumed to be understated in economic statistics. Controversy exists about the best method for estimating the extent of under-reporting and about the resulting measures of the size of the underground economy. This paper refines a method developed by Pissarides and Weber (1989) and uses discrepancies between food shares and reported incomes of the self-employed and other households to estimate under-reporting by the self-employed. In contrast to previous studies our panel data methodology distinguishes income under-reporting from transitory income fluctuations of the selfemployed, and provides an exact estimate of the degree of underreporting rather than just an interval estimate. Using panel data from Korea and Russia we estimate that 38 percent of the income of selfemployed households in Korea and 47 percent of the income of Russian self-employed households is not reported.