Diets with less protein improve profitability unless the savings from feeding less protein are outweighed by lost milk revenue. Our objective was to quantify the variation among cows in their ability to maintain production (be resilient) when fed protein-deficient diets. Mid-lactation Holstein cows with initial energy-corrected milk (ECM) of 40 ± 9 kg/d (n = 149, in 5 blocks) were fed either a high protein diet (18% CP; HP) or low protein diet (14%CP; LP) in a crossover design with 2 treatment periods of at least 28 d/period. Both diets contained at least 9.8% rumen-degraded protein. HP and LP diets were similar for each block, but HP contained expeller soybean meal that was replaced by soybean hulls and ground corn in LP. HP contained 4% units more CP, 2% units less starch, and 2% units less NDF than LP. Cows were fed ad libitum. Fixed effects of block, parity, diet, period, cohort, and cohort by diet, and random effect of cow were included in the model to examine milk output and total capture of energy and protein (assuming body energy gain = 6 × BW gain/d and body protein gain = 0.14 × BW gain/d) between diets. Compared with LP, HP increased ECM (39.4 vs. 35.4 kg/d; P < 0.01), milk protein (1.28 vs. 1.14 kg/d; P < 0.01), BW gain (0.47 vs. 0.11 kg/d; P < 0.01), MUN (15.1 vs. 9.3 mg/dL; P < 0.01), captured energy (32.4 vs. 27.2 Mcal/d; P < 0.01), and captured protein (1.35 vs. 1.15 kg/d; P < 0.01). Decreases in milk and captured energy and protein for each cow when fed LP compared with HP were calculated and modeled by ECM per kg metabolic BW when fed HP, DIM, parity, cohort, block, and all interactions. The residual term from the prediction model was used to quantify variation among cows for resilience to LP. The R², means, and standard deviations (SD) of means and residuals were, respectively, (1) 0.74, 4.0, 3.2 and 1.6 for kg ECM; (2) 0.80, 0.14, 0.10, and 0.05 for kg milk protein; (3) 0.60, 5.0, 5.5 and 3.5 for Mcal captured energy; and (4) 0.67, 0.17, 0.16 and 0.09 for kg captured protein. We conclude that half of the variation remains for the residual term and suggest that genetics contributes to part of this variation.