Subclinical ketosis is a metabolic disorder in high-producing dairy cattle that can be detected by ketone bodies in milk: acetone (Ac), acetoacetate (AcAc), and beta-hydroxybutyrate (BHBA). Fourier transform infrared (FTIR) spectrometry is to a growing extent used for determination of milk constituents in milk recording, but as yet there is no calibration for ketone bodies available. The objective of this study was therefore to build a calibration for the MilkoScan FT6000 (FOSS Analytical A/S, Hillerød, Denmark) for Ac, AcAc, and BHBA and to evaluate the FTIR predictions for detection of subclinical ketosis. From 217 herds, 1,080 milk samples were taken from fresh multiparous dairy cows. The Ac, AcAc, and BHBA concentrations were determined by chemical methods using segmented flow analysis. Because of its low concentration, AcAc seemed to be hardly detectable and was therefore not considered further. The correlation between the chemical method results of Ac and BHBA was 0.82, indicating that both ketone bodies were elevated in milk during subclinical ketosis. In wk 1 postpartum, however, most samples with a high Ac concentration did not have a high BHBA concentration, whereas after wk 5 postpartum most samples with a high BHBA concentration did not have a high Ac concentration. For Ac and BHBA, the correlation coefficients between the FTIR predictions and the chemical results were around 0.80 with standard error of cross validation values of 0.184 and 0.064 mM for Ac and BHBA, respectively. Using thresholds of 0.15 mM for Ac and 0.10 mM for BHBA, high values for Ac or BHBA were detected with a sensitivity of 69 to 70%, a specificity of 95%, with 25 to 27% false positives and 6 to 7% false negatives. It is argued that FTIR predictions for Ac and BHBA are valuable for screening cows on subclinical ketosis, especially when used in combination with other indicators, and can serve in the evaluation of the herd health status with respect to subclinical ketosis.