We develop a bottom-up approach to estimating the slope of the primitive form of the New Keynesian Phillips curve, which features marginal cost as the relevant real activity variable. Using quarterly micro data on prices, costs, and output from the Belgian manufacturing sector, we estimate dynamic pass-through regressions that identify the degrees of nominal and real rigidities in price setting. Our estimates imply a high slope for the marginal cost-based Phillips curve, which contrasts with the low estimates of the conventional unemployment or output-based formulations in the literature. We reconcile the difference by demonstrating that, although the pass-through of marginal cost into inflation is substantial, the elasticity of marginal cost with respect to the output gap is low, at least for pre-pandemic data. We also illustrate the advantage of a marginal cost-based Phillips curve for characterizing the transmission of supply shocks to inflation.