There are long-standing concerns that household income mobility is over-estimated due to measurement errors in reported incomes, especially in developing countries where collecting reliable survey data is often difficult. We propose a new approach that exploits the existence of three waves of panel data to can be used to simultaneously estimate the extent of income mobility and the reliability of the income measure. This estimator is more efficient than 2SLS estimators used in other studies and produces over-identifying restrictions that can be used to test the validity of our identifying assumptions. We also introduce a nonparametric generalisation in which both the speed of income convergence and the reliability of the income measure varies with the initial income level. This approach is applied to a three-wave South African panel dataset. The results suggest that the conventional method over-estimates the extent of income mobility by a factor of more than 4 and that about 20% of variation in reported household income is due to measurement error. This result is robust to the choice of income mobility measure. Nonparametric estimates show that there is relatively high (upward) income mobility for poor households, but very little (downward) income mobility for rich households, and that income is much more reliably captured for rich than for poor households.