The Cornell Net Carbohydrate and Protein System (CNCPS) adopts energy requirements as Meals of metabolizable energy (ME) largely as NRC (1988), which are to be used with feed ME values as measured at the maintenance level of feeding. However, the model inserts calculated production level ME values into these NRC standards, equivalent to an upwards correction of about 5% to the estimates of ME requirements adopted. The energy accounting of the model is therefore flawed. The maintenance requirement of all breeds of dairy cattle, other than Holstein, are increased by a factor of 1.2, based on work with beef suckler cows. The efficiency of ME use for milk synthesis is also raised to a constant 0.65 from the normal range of 0.62 to 0.64 specified in NRC (1988). The net effect on the adopted ME requirements at milk yields of 30 kg/d is small, but both slope and intercept of the equation differ from NRC (1988). The model includes no effect of diet amount and composition upon nutrient partitioning between milk and body. Neither is there any effect of diet composition upon predicted milk composition, which is either an input to the model or a function of day of lactation. The body composition of growing heifers and cows up to four years of age is predicted by a function which has a maximum body fat content of 22.5%, only 0.86 of the total body fat recorded in recent body composition measurements in Friesian dairy cows. Associated estimates of maximum mobilizable body fat are one half or more below recent measurements with dairy cows. The handling of energy losses and gains during lactation uses condition score as a measure and ignores liveweight change. This is also based on research with suckler beef cows, and shown to over-estimate the energy equivalent of a unit change in condition score of Holstein dairy cows. Prediction of dry matter intakes is closer to actual than other prediction functions available and the lag in intake in early lactation is also accommodated well. The consequence of using mismatched energy requirements (as ME) is that the model predicts significantly lower milk yields at zero energy balance (or lower energy 362 CORNELL SYSTEM FOR DAIRY CATTLE balances if milk yield is given) compared to both NRC (1988) and AFRC (1993). However, field tests of the model have shown that predicted milk yields are closer to actual than those from other models.