The aim of this study was to evaluate predictive models of the chemical composition of kikuyo grass (Pennisetum clandestinum) using near infrared spectroscopy (NIRS). Samples of P. clandestinum were collected from the district of Florida-Pomacochas, Amazonas, Peru, in three stages of the plant (45, 60 and 75 days after cutting) and two seasons (rainy and dry). The moisture content (H), crude protein (PC), ether extract (EE), crude fibre (FC), ash and gross energy (EB) were determined. The absorbance spectra were obtained in the wavelength range of 1100-2500 nm. Through Matlab 2015ª functions and scripts, complete and optimized prediction models were implemented using neural networks (ANN) and regression by partial least squares (PLSR). The optimized models used 18 relevant wavelengths, determined for both types of models according to the matrix of beta coefficients of the PLSR model. The models PLSR vs ANN, in the validation stage, showed better fit (R2>0.70) in H, PC, EE, ash and EB with an R2 of 0.74, 0.89, 0.79, 0.74 and 0.87, respectively. Therefore, the NIRS-PLSR model has potential in the prediction of the composition of the kikuyo grass (P. clandestinum).