Near Infrared Spectroscopy (NIRS) is widely used for the assessment of animal feed composition, including compound feeds, raw materials, forages and total mixed rations (TMR). Usually, these composition data are incorporated intofeeding equations or models to predict the animal response. However, a NIR spectrum is a powerful digital print of the product, containing not only information about the chemical composition of the product, but also information about other parameters related to its texture, origin, structure or organoleptic characteristics. Thus, the objective of this work is to evaluate the potential of NIRS to directly predict the voluntary intake of dairy cows from TMR spectra. Sixty cows belonging to Aarhus University farm were divided in four groups and fed with four different TMR diets (control diet, and three other diets with an increasing level of 5, 10 and 15% of straw instead of silage). Cows were monitored during eight weeks and feed intake was measured individually. A total of 142 samples of TMR were analysed using a portable micro-NIRS instrument, based on LFV (Linear Filter Variable) technology, working on reflectance in therange 910-1676 nm. A PLS (Partial Least Squares) model was developed for predicting the total TMR intake with a population characterized by a total feed intake mean value of 49.52 kg, resulting a coefficient of determination of0.81 and a standard error of cross validation (SECV) value of 2.52 kg. These results show the feasibility of NIRS for the in situ direct prediction of the animal response associated to a specific diet, which could be of great importanceto match animal needs and diets, and to increase the farm efficiency.