Improving feed efficiency is expected to reduce resource use while maintaining production. This achievement depends on the ability to identify the most and least efficient animals. The objective of the current study was to analyze the impact of measurement period length and start point during lactation on the ability to identify the most and the least efficient dairy cows. Individual daily feed intake, milk net energy, metabolic BW, and BCS were monitored frequently during the first 224 DIM for 117 Holstein cows fed a same total mixed ration. Different period durations and start points were defined: duration went from 14 to 196 consecutive days and start points went from DIM 1 to 210 with 14-d increments. Feed efficiency was estimated with the residual energy intake (REI) as the residuals of a multiple linear regression estimating average net energy intake from average milk net energy, average metabolic BW, and body condition score loss and gain. This model was adjusted over the 224 d, the reference period in this study, and within each short period. Ability to identify the most and least efficient cows was estimated with the positive predicted value (PPV) and the sensitivity. PPV is the percentage of most and least efficient cows identified within a short period, which are also identified as such over the reference period. Sensitivity is the percentage of most and least efficient cows over the reference period, which are also identified as such within a short period. The most and least efficient cows were the cows outside the range of REI average plus and minus one standard deviation. Measurement start point had a positive linear effect on PPV (P < 0.01), but not on sensitivity (P = 0.12). Period duration had a positive linear effect (P < 0.01) and a negative quadratic effect (P < 0.01) on both PPV and sensitivity, highlighting that PVV and sensitivity reached a maximum. A minimum of 112 d were necessary to achieve a sensitivity and a PPV, which were not significantly different (P > 0.06) from their respective maximum: 86% and 79%. To decrease the risk of identifying efficient cows as inefficient or inefficient cows as efficient, REI should be estimated over at least 112 d to identify a maximum of the efficient and inefficient cows.