Among horse breeders, distinctive facial features have been anecdotally attributed to traits such as trainability and soundness. Genes controlling early hormonal development have been implicated but never formally proven to drive correlations between facial morphology, and health in human and animal literature. A pilot study demonstrated significant associations between health PTAs and facial biometric measurements among diary sires. In this study, side profile images were acquired from 573 genotyped Holstein cows of mixed parity, of which 344 also had genomic estimates for wellness traits from Clarifide Plus (Zoetis, Service LLC). Images were manually annotated with anatomical landmarks, and 60 validated biometric measures computed. Biometric values were then combined with genomic estimates of type traits in independently optimized statistical learning models to predict the 22 available genetic merit estimates for performance, fertility, and health. LASSO regression models, fit by k-fold cross validation using the glmnet package in R, revealed little improvement in R² values on the validation set with addition of image data, and yet a subset of biometrics was retained in over half the models. Further analysis using smoothing splines implemented in the mgcv package revealed evidence of nonlinearity in these relationships, with a notable number of significant associations to calving ease and stillbirth values. Finally, bagged regression trees optimized by k-fold cross validation on a custom grid in the gbm package showed little evidence of significant interaction effects and little improvement in R² on the validation set, though higher depth trees for retained placenta and metritis did show an improvement of 0.35 R² over type traits alone in their respective training models. Variable importance measures also revealed biometric values to be significant components across the range of models. While biometrics did not consistently improve predictions of genetic merit over type traits, their consistent inclusion in cross-validated models and low phenotypic correlation to type traits (mean 0.05, max 0.20) suggests they may provide novel useful information to dairy genetic evaluations as potential indicator traits.