Measuring occupational mobility from the Current Population Survey using recall (retrospective) or linked panel responses (longitudinal) generates substantially different outcomes, both in levels and trends. Using a generalized method of moments technique, we estimate the actual level of occupational mobility and the measurement error in both of these measures for 1981-2018. Measurement error in longitudinal measures is large and has been worsening over time. However, actual occupational mobility is approximately 70% higher than retrospectively measures. Our estimated corrections imply workers in tradable occupations are less likely to switch occupations than previously believed, implying potentially lower welfare gains from trade.