This study evaluated the Cornell Net Carbohydrate and Protein System for dairy cows consuming diets based on pasture, assessed the sensitivity of the model to critical inputs, and demonstrated application opportunities. Data were obtained from four grazing experiments and four indoor pasture feeding experiments (25 dietary treatments) involving dairy cows in New Zealand and the US. The model provided a reasonably good estimate of changes in body condition score (r2 = 0.78; slope not significantly different from 1), estimated energy balance (r2 = 0.76; slope not significantly different from 1), blood urea N (r2 = 0.94; underprediction bias of 0.5%), microbial N flow (r2 = 0.88; slope not significantly different from 1), and milk production. The model underpredicted dry matter intake (r2 = 0.80; 13% bias) and overpredicted ruminal pH (r2 = 0.47; 1.7% bias). Predicted milk production was especially sensitive to changes in pasture lignin content, effective fiber, rate of fiber digestion, and amino acid composition of ruminal microbes. Milk production was first-limited by the supply of metabolizable energy when only high quality pasture was fed, but specific amino acids limited milk production when more than 20% of the diet consisted of a grain supplement. These results indicate that the Cornell Net Carbohydrate and Protein System can be used for dairy cows in a grazing system to make realistic predictions of performance.