Negative energy status in early lactation is linked to a variety of metabolic disorders, reduced fertility, and decreased milk production. To improve the energy status of cows by breeding and management, the identification of negative energy status is crucial. While biomarkers such as non-esterified fatty acid (NEFA) concentration and beta-hydroxybutyrate (BHB) in blood plasma could be used to identify a negative energy state, measuring them directly from blood is both invasive and expensive. In this work, we developed prediction equations for blood plasma NEFA and BHB levels based on mid-IR spectral measurements of milk. The models were fitted using partial least squares regression and evaluated using both cross-validation and independent-herd validation. A total of 3183 spectral records from 606 lactations originating from three different herds were utilised. R2 values of 0.53 (RMSE=0.206mmol/l, RMSE of cross-validation (RMSECV) 0.217mmol/l) for NEFA and 0.63 (RMSE=0.326mmol/l, RMSECV=0.353mmol/l) for BHB were obtained. Furthermore, relatively similar prediction accuracies were found for BHB (RMSE of prediction (RMSEP) 0.411mmol/l and 0.422mmol/l) and NEFA (RMSEP=0.186mmol/l and 0.221mmol/l) when model training was done using two herds and validated on the third herd. The results from the model fits confirm that it is possible to build blood plasma BHB and NEFA models based on mid-IR spectra that are sufficiently accurate for practical use.