Equations to predict dry matter intake (DMI) of crossbred Holstein-Zebu dairy cows were developed and compared by using data of 161 treatment means from 38 Brazilian studies (n = 446 cows, milk production average = 16.60 ± 5.70 kg/d). A data set was developed of Holstein × Zebu lactating dairy cows of different degrees of breeding in confinement or pasture systems. The data were evaluated using mixed nonlinear models including study as a random effect. Body weight (BW), 4% fat corrected milk (4% FCM) and weeks of lactation (WOL) were used as independent variables in the model. The proposed model to estimate DMI (kg/d) of crossbred cows was: [0.5552 (±0.06636) × 4% FCM + 0.06332 (±0.009455) × BW^0.75] × [1 – e^{(−0.7732 (±0.7019) × (WOL - 1.629 (±1.913)))}]. The accuracy of the model was compared with 4 previously published equations by use of an independent data set. The mean square error of prediction (MSEP), mean bias, concordance correlation coefficient (CCC), and analysis of linear regression were used for evaluating models. The new model showed the lowest values of MSEP and highest CCC and r² compared with the other 4 equations (Table 1). The Brazilian model from Freitas et al. (2006), despite being developed in tropical conditions, showed the highest value of the MSEP. The new equations to predict DMI can be used in the formulation of diets for crossbred dairy cows under tropical conditions. Table 1.

Variable	DMI(kg/d)	New model(2015)	NRC(2001)	Fox et al.(2004)	Traxler(1997)	Freitas et al.(2006)
Mean, kg	16.10	16.88	16.11	15.05	16.95	15.85
Mean bias, Y – X, kg	—	−0.78	−0.01	1.05	−0.85	0.25
MSEP, kg × kg	—	1.64	2.45	2.73	2.76	6.98
MSEP Decomposition, %
Mean bias, %	—	37.001	0.003	40.458	26.338	0.863
Systematic bias, %	—	3.469	0.052	5.327	0.781	63.86
Random errors, %	—	59.531	99.945	54.215	72.882	35.227
CCC	—	0.90	0.82	0.80	0.80	0.75
r2	—	0.88	0.69	0.81	0.74	0.69
P-value (a = 0)	—	0.495	0.889	0.229	0.244	<0.001
P-value (b = 1)	—	0.125	0.884	0.049	0.506	<0.001