Maize silage nutritive quality is routinely determined by near infrared reflectance spectroscopy
(NIRS). However, little is known about the impact of sample preparation on the accuracy of the
calibration to predict biological traits. A sample population of 48 maize silages representing a wide
range of physiological maturities was used in a study to determine the impact of different sample
preparation procedures (i.e., drying regimes; the presence or absence of residual moisture; the degree
of particle comminution) on resultant NIR prediction statistics. All silages were scanned using a total
of 12 combinations of sample pre-treatments. Each sample preparation combination was subjected to
three multivariate regression techniques to give a total of 36 predictions per biological trait. Increased
sample preparations procedure, relative to scanning the unprocessed whole plant (WP) material,
always resulted in a numerical minimisation of model statistics. However, the ability of each of the
treatments to significantly minimise the model statistics differed. Particle comminution was the most