A cane quality evaluation system using near infrared (NIR) spectroscopy for determining payments to growers was developed and intro- duced to raw sugar mills in southern Japan. The evaluation system consists of a cutter grinder for sample preparation, an NIR instru- ment and a networking system. We investigated a calibration model for the determination of sugar index [Pol in cane, (PIC)] for shredded cane samples using a modi ed partial least squares regression with and without a repeatability le. The accuracy of the model devel- oped using the repeatability le was adequate [root-mean square error of prediction, (RMSEP)=0.30%] and the pooled standard error (P-SE) between master and eight slave instruments was 0.14%. In contrast, calibration models without a repeatability le incurred large P-SE and pooled bias (P-bias) values, although the RMSEP values were very similar to the repeatability le model in cases using the rst derivative and standard normal variate (SNV) spectral pre-treatment. The evaluation system was installed in 17 raw sugar mills on 13 islands involved in sugarcane cultivation in southern Japan. The system is presently used to measure the PIC value of cane samples that are supplied by farmers to mills. All data obtained are collected and compiled into a central facility via the Internet. The NIR networking system for sugarcane has been implemented in all raw sugar factories in Japan as an of cial method for determining cane prices.