Residual feed intake (RFI) is an inheritable indication of feed efficiency that is independent on level of production. Many factors may affect the variation of RFI, but the physiological and metabolic mechanisms underlying RFI are not fully elucidated. This study was conducted to investigate dietary nitrogen (N) partitioning and microbial protein synthesis in lactating dairy cows selected for different feed efficiency. Thirty multiparous Holstein dairy cows, with milk yield of 35.3 (±4.71 SD) kg/d, were used to derive RFI. Milk yields from all cows were recorded at each milking, and milk samples were collected and analyzed per week. Spot urine samples were taken on d 24–25 and d 49–50 before daily feeding. Urinary purine derivatives were used to indirectly estimate the microbial protein flow in the rumen. All the data were analyzed using PROC MIXED of SAS. After the RFI measurement period of 50 d, the 10 lowest RFI cows (−0.96 kg/d ± 0.34 SD) and 8 highest RFI cows (1.18 kg/d ± 0.28 SD) were selected. The lower RFI cows (more efficient) had 2.45 kg/d less dry matter intake (DMI) than the higher RFI cows (P < 0.05), but they produced similar energy-corrected milk. The ratios of milk to DMI (1.41 vs.1.24, P < 0.01) and energy-corrected milk to DMI (1.48 vs.1.36, P < 0.01) were higher in the lower RFI cows than those in the higher RFI cows. Milk urea nitrogen was lower in the lower RFI cows than in the higher RFI cows (12.7 vs. 13.9 mg/dL, P < 0.05). Compared with higher RFI animals, the lower RFI cows had a higher partition of feed N to milk N (29.7 vs. 26.5%, P < 0.05). Numerically, though not significantly, higher ratios of microbial protein to rumen degradable protein and metabolizable protein to crude protein intake were obtained in the lower RFI cows compared with higher RFI cows, eventually resulting in higher proportion of milk protein to the dietary CP. In summary, all processes from intake to milk protein secretion, including N partition, synthesis of microbial protein, and utilization of metabolizable protein, contribute to the variance in the RFI.