This study explores the measurement errors embedded in the income data of the Regional Employment Survey (RES) using earnings data of National Tax Services (NTS) by region, sex and percentiles. This study mainly presents three findings. First, the surveyed income data have mean-reverting measurement errors. Second, the errors embedded in the dependent variable of the surveyed income data attenuate the coefficients in bi-variate regression model. Third, they are serially auto-correlated. The errors are mostly eliminated by a first-difference ordinary least square model. The findings suggest the need to confirm a conventional assumption of no error or classical measurement error made in empirical studies.