Inferring composition of dairy cattle milk fatty acids (FA) is of interest for both human nutrition and potentially for dairy management. Fourier-transform infrared spectroscopy (FTIR) has been used to predict milk composition for payment on US dairy farms. Our study aimed to assess prediction abilities for FA content with several models using FTIR data. Milk samples were drawn from 365 early lactation (<90 DIM) cows on 4 Michigan dairy farms. We used FTIR absorbance information on 899 different wavelengths in the mid infrared region generated from 2 different Bentley FTS/FCM NEXGEN spectrometers at NorthStar Cooperative (Grand Ledge, MI) to predict the content of 7 saturated FA (C4:0, C6:0, C10:0, C12:0, C14:0, C16:0, C18:0), 3 unsaturated FA (C14:1 9c, C18:1 9c, C18:2 9c, 12c), and the overall total of saturated (SFA), monounsaturated FA (MUFA) and poly-unsaturated FA (PUFA) all determined using gas liquid chromatography. Content of FA was expressed in g per 100 g of milk. The models tested included partial least squares regression with 5, 10, 15 and 20 components (PLSR5, PLSR10, PLSR15, PLSR20), and 2 Bayesian regression models based on heavy tailed (BayesA) or variable selection (BayesB) specifications. Predictions were assessed by the correlation (COR) between predicted and actual values using a leave-one-herd-out cross-validation approach. The lowest COR was observed for C14:1 9c (~0.36), C10 and C12 (~0.47). The highest COR obtained was with SFA (0.93) and C16 (0.88). For SFA, COR varied between validation herds from 0.87 (PLSR20) to 0.97 (all models except PLSR20); for MUFA, from 0.50 (PLSR20) to 0.92 (BayesA/B) whereas for PUFA, varied from 0.55 (PLSR20) to 0.89 (BayesA/B). We conclude that prediction of FA by FTIR is feasible although predictive performance varied for different FA and for different herds. PLSR models and Bayesian models showed similar prediction performance; however, PLSR model predictions dramatically depended upon the number of components considered. Further developments will be crucial to better utilize widely available FTIR data to predict fine FA composition of milk for human nutrition, estimate cow genetic merit for FA, and for dairy management applications