Objectives of this study were to characterize feeding-behavior (FB) patterns in growing dairy heifers with divergent RFI phenotypes (0.50 SD) and to evaluate the accuracy of partial least squares regression (PLSR) models to predict RFI based on FB traits. Performance, DMI, and FB traits were measured for 70 to 100 d in 15 trials with Holstein heifers (n = 611) fed a corn-silage based ration. Seventeen FB traits were evaluated: frequency and duration of bunk visit (BV) and meal events, head-down duration (HDD), meal length, maximum non-feeding interval, corresponding day-to-day variation (SD) of these traits, and ratios of HDD per BV duration and meal duration, HDD per meal duration, and BV events per meal event. Data were analyzed using a mixed model that included RFI group and trial. The PLSR model for RFI was developed using cross-validation procedures (Leave-One-Out) in JMP (SAS), with FB traits as independent variables. Low-RFI heifers consumed 24% less (P < 0.01) DMI and had lower (P < 0.01) day-to-day DMI variation than high-RFI heifers. Distinct differences were observed in FB patterns between low- and high-RFI heifers (Table 1). Eight of 17 FB traits were included [selected based on variable of importance (VIP) score > 0.80] in the PLSR model that explained 33% of the variation in RFI. Head-down duration had the highest VIP score; accordingly, low-RFI animals had 44% lower HDD and 30 and 40% lower ratios of HDD per BV duration and meal duration, respectively. Additionally, low-RFI animals had 20 and 18% fewer BV and meal events per day, spent 21% less time eating during BV events, and had reduced day-to-day variation in HDD and meal frequency. For this study, distinctive differences were observed in the FB patterns of Holstein heifers with divergent RFI, which explained 33% of the between-animal variation in RFI.