Feed conversion efficiency is important for the profitability of dairy farms, but measuring dry matter intake to estimate it directly is expensive and time consuming. Consequently, fast and inexpensive measurements of genetically correlated traits are desirable as predictors. In this study, we describe the utility of thermal imaging to predict feed conversion efficiency. We conducted a 30–40 d feeding trial on 6 to 9 mo old Friesian bulls (n = 75) and their half-sisters (n = 246) and estimated feed conversion efficiency as residual feed intake (RFI: the residual from a regression of daily dry matter intake on average daily gain and mid-trial metabolic weight). We also took thermal images to measure heat loss from the eye, cheek, and muzzle. Using univariate animal models, we estimated the heritability of RFI for each sex (0.13 for heifers and 0.18 for bulls), and using a bivariate animal model that treated RFI in bulls and heifers as different traits, we estimated the between-sex genetic correlation (0.93). To investigate the utility of thermal traits as predictors, we fit single-trait animal models for all 8 heat loss measurements (maximum and mean temperature of the eye, eye corner, cheek, and muzzle) to estimate heat loss EBVs and then used multiple regression of heat loss EBVS on RFI EBVs to produce prediction equations. Even over-fit regression models using all 8 predictors resulted in low coefficients of determination (r²) of 0.35, 0.40, and 0.19 for both sexes, heifers and bulls respectively. To simulate selective RFI phenotyping of bulls only, we fit a 3-trait animal model treating RFI and the first 2 principle components (PCs) of the 8 heat loss traits as separate traits using all data on bulls, but only the heat loss PCs on heifers to estimate RFI BVs in both sexes. These RFI EBVS estimated using only 75 direct measurements on bulls were highly correlated with those from univariate models using 321 measurements on both sexes with r² values of 0.85, 0.84, and 0.94 for both sexes, heifers and bulls respectively but with lower reliabilities. Both strategies would require extensive independent validation to justify routine measurement of RFI and its incorporation into the national breeding objective.