This thesis examines the capability of two different MicroNIR spectrometers to measure the composition of raw milk for an online milk sensor. On-farm milk testing is an important tool in a developed dairy market. It is used by farmers to determine milk quality, nutrition and individual animal performance. The price farmers receive for milk is based on the fat and protein content. Currently milk composition is monitored on average three times a season, which leads to late diagnosis of poor performance. Daily monitoring of milk composition allows farmers better farm management and optimisation of the processes and feed to achieve success. The International Committee for Animal Recording (ICAR) sets standards for sensors which give farmers confidence in the data they receive. Two low cost spectrometers were assessed to see if they meet the ICAR fat and protein regulations for an online sensor. The spectrometers covered two spectral regions in the NIR range and were both developed by JDS Uniphase Corporation (JDSU, California, USA). The MicroNIR1700 spectrometer analyses the 870 – 1660 nm region, and the MicroNIR2200 spectrometer covers the 1130 – 2150 nm region. An initial laboratory trial with the unmodified spectrometers did not meet ICAR precision limits for milk analysis. The high prediction errors were caused by insufficient signal, aged milk samples being reheated and excessive dark current. To improve the signal to noise ratio of the spectrum the lamp voltage was increased. This increased the received signal to the spectrometers but also increased the temperature of the spectrometers. The increase in temperature caused the dark current to excessively increase. To reduce the operating temperature of the spectrometers a thermoelectric cooler was used to cool the spectrometers to 20°C. The modified spectrometers with increased light and thermoelectric cooler were tested on three different farms to establish if the modifications done can meet ICAR fat and protein precision limits. A total of 590 milk samples were collected across three different farms. Calibration models were tested on each individual farm and across all three farms to assess the capability of calibration models to be transferred across the different farms. Off-line and on-farm, the modified MicroNIR spectrometers achieved ICAR fat and protein precision limits for an online sensor. For the pooled data set the MicroNIR1700 spectrometer predicted fat and protein with a root mean square error of 0.061 and 0.151 respectively. The MicroNIR2200 predicted fat and protein with a RMSE of 0.061 and 0.138. The independent calibration models could be transferred across farms with only a slight increase in prediction error but still within the ICAR tolerances for an online milk sensor It is concluded the original MicroNIR spectrometers could not predict fat and protein to ICAR precision limits for an online sensor. To achieve ICAR accuracy with the MicroNIR spectrometers the light output needed to be increased and cooling was required to reduce the operating temperature. This work shows that there is definite potential for low cost spectrometers to be used for the determination of milk properties.