This paper studies the dynamics of labor demand at the plant level to quantify labor adjustment costs. At the plant level, in contrast to time-series observations, the correlation of hours and employment growth is negative while hours and employment growth are about equally volatile. We specify and estimate the parameters of a plant-level dynamic optimization problem using simulated method of moments to match these plant-level observations. Our findings indicate that non-convex adjustment costs are critical for explaining plant-level observations on hours and employment adjustment.