Abstract: A prediction model was built using the technology of Near Infrared Spectroscopy (NIRS) with fiber optic probe and method of partial least squares regression (PLS ) . Multiplicative Signal Correction (MSC ) , Standard Normal Variate, First Derivative and Smoothing were used to pretreat the spectra. The best effect, with a correlation coefficient of 0.9698 were achieved by the petreatments of MSC full, First Derivative BCAP and smooth average 9 points, a standard analytical error of calibration (SEC) of 0.1131% and a RPD of 6.01 for the calibration sample set and a correlation coefficient of 0.9402, a standard analytical error of prediction (SEP) of 0.1536% and a RPD of 4.04 for the validation sample set. The result showed that the technology of Near Infrared Spectroscopy (NIRS) with fiber optic probe could be used to protect crud fiber content in forage.