Ash content is a useful parameter in forage and grass quality studies, but the traditional methods used to quantify this parameter are tedious and sample processing time-consumig is. Near infrared reflectance spectroscopy (NIRS) overcomes some drawbacks of the traditional methods. Measurement of a sample's diffuse reflectance is rapid and non-destructive and chemical reagents are not necessary, which is environmentally friendly. We predicted the ash content of herbage samples from semiarid grassland communities by NIRS. The samples were collected on different sites, over four consecutive years. The calibration equations generated from log 1/reflectance to predict ash content had a standard error of calibration (SEC) of 4.6 g kg-1 and R2 of 0.88. The standard error of prediction (SEP) was 5.1 g kg-1 and r was 0.94.