This note estimates the long-run demand for skilled and unskilled labor using panel data for Colombian manufacturing plants. Unobserved heterogeneity and measurement error problems that commonly arise in microdata production estimates are examined. Output measurement errors cause OLS estimators to underestimate the output and wage response of employment demand. Time-difference estimators exaggerate the measurement error biases. Instrumental-variable estimates of the output elasticities are 0.89 and 0.76 and own-wage elasticities are −0.42 and −0.65 for skilled and unskilled labor, respectively. The output elasticity is larger for skilled labor whereas the wage elasticity is larger for unskilled labor in virtually every industry.