I report the measurement error in self-reported earnings for a developing country using a novel data set. The data set consists of two cross-sections of the Federated States of Micronesia (FSM) wage and salary sectors; additionally, a subset of the two cross-sections may be linked to create a panel. Administrative data from FSM Social Security office are matched to the FSM Census data for the wage and salary sectors. I find that the error in annual self-reported earnings is centered on zero. Additionally, I find strong evidence for mean reversion in the data suggesting non-classical measurement error. I identify the impact of prior years' earnings variability on the current reporting of earnings using administrative data on earnings histories. Prior earnings volatility strongly affects measurement error in current period. However, the effect of prior shocks diminish significantly over time--suggesting that first-differencing and fixed-effects techniques will not improve accuracy.