Earlier studies showed the superiority of NIRS, over fibre measurements, for predicting organic matter digestibility (OMD) in vivo of grass silages. (Barber et al 1990). This system was put into routine use in ADAS in 1989 and after some initial doubts, due to the wider range of the predicted data seen, is now accepted as the best system available for routine use. However, occasional erroneous values were predicted for which there were no obvious explanations, and which resulted in occasional relatively poor repeatability. In common with all NIRS applications it was likely that the sources of the problems contributing to the errors were i) instrumental and environmental noise, ii) sample particle size effects and iii) variable moisture content of the samples. A course of investigation was undertaken with the objective of determining the effects of these on the predicted data.