The objective of this experiment was to develop and evaluate models predicting hyperketonemia (HYK), utilizing several methods and variable inputs. Paired blood and milk samples were collected from multiparous cows 5 to 18 d in milk at 3 WI farms (3,629 observations from 1,013 cows). Blood β-hydroxybutyrate (BHB) concentration was determined by the Precision Xtra meter and milk samples were analyzed by a commercial lab (AgSource) for components and mid-infrared spectrum absorbance. Cow specific variables were extracted from DairyComp 305. A BHB ≥ 1.2 mM was considered HYK, resulting in a prevalence of 12.4%. The data set was divided into an external testing set (n = 609) and a training data set (n = 3020). Model fitting was completed with JMP 12. A 5-fold cross-validation (CV) was performed on the training data set for 3 methods with square root of BHB as the model dependent: multivariate linear regression (MLR), partial least squares regression (PLS) and artificial neural network (ANN). Each method was fitted utilizing 3 combinations of potential variables: milk spectrum, management variables (milk components and DairyComp 305 data), or all variables, resulting in 9 models. All models were evaluated based on r², RMSE, and the area under the curve (AUC) of a receiver operating characteristic curve. Data are expressed as the mean ± SE of CV. Across modeling techniques, use of all data resulted in greater performing models than use of management variables (intermediary) or spectrum data. All MLR models were lower performing than other methods and MLR-spectrum was the lowest with an r², RMSE, and AUC of 0.23 ± 0.02, 0.45 ± 0.07, and 0.81 ± 0.01, respectively. PLS models performed intermediately with PLS-all values: 0.40 ± 0.01, 0.17 ± 0.002, 0.86 ± 0.01 for r², RMSE, and AUC. The ANN method performed the greatest, particularly the ANN-all with an r², RMSE, and AUC of 0.46 ± 0.01, 0.16 ± 0.002, and 0.86 ± 0.004. When the ANN-all model was validated against the test data set, r² = 0.56, RMSE = 0.16, sensitivity was 83%, and specificity was 80%. In summary, use of ANN methods with milk spectrum and management variables can achieve reasonable prediction of HYK.