Subclinical ketosis affects between 40 and 60% of dairy cows and negatively affects cow productivity and health. Although cowside ketone testing strategies are available, many lack sufficient accuracy, are labor-intensive, and can be costly. The objective of this study was to validate the use of multiple regression models to predict blood β-hydroxybutyrate (BHBA) from milk components and continuous test-day variables in early lactation cows for determining ketosis prevalence. Blood samples were collected on the same day as milk test from 658 Holstein cows 5 to 20 DIM on 10 dairy farms. Blood serum was analyzed for concentration of BHBA by colorimetric assay (Stanbio, Boerne, TX). Milk samples were analyzed for milk BHBA and acetone concentrations by Fourier transform infrared spectrometry (FOSS Analytical A/S, Eden Prairie, MN), in addition to standard milk analysis variables. Continuous test-day variables were collected from DairyComp305 (Valley Agricultural Software, Tulare, CA) records. Models were built in the REG procedure of SAS 9.4 (SAS Institute Inc., Cary, N.C.) using stepwise, forward selection by excluding variables with a P-value < 0.15 and selection criterion of Akaike’s information criterion. Data interrogation justified development of separate models for primiparous and multiparous cows, as well as cows 5 to 11 DIM and cows 12 to 20 DIM. Additionally, disease etiology allowed for unique models for the 5 to 11 and 12 to 20 DIM ranges. Significant variables were milk BHBA, acetone, and fat:protein ratio; parity, previous days dry, previous lactation length, and age at first calving; and DIM and milk production on test day. Overall, model accuracy was 88% for multiparous cows 5 to 11 DIM (R² = 0.57), 83% for multiparous cows 12 to 20 DIM (R² = 0.67), 96% for primiparous cows 5 to 11 DIM (R² = 0.74), and 97% for primiparous cows 5 to 20 DIM (R² = 0.66). These results suggest that modeling blood BHBA based on milk component data and cow test-day information can serve as a valuable diagnostic tool for monitoring herd-level ketosis prevalence.