Typical current protocols require lamps 1-2 h to stabilise before initial white reference measurements on NIRS instruments. However many in-line analytical applications that employ a near infrared (NIR) spectrometric system, which involves scanning of numerous samples each second, making reference measurements difficult and inconvenient. Therefore, in these cases most take references only periodically, often after extended periods. Further, in many industry applications, ambient temperatures in many industry applications can vary by 50 °C or more. As the spectral output of the illuminating source, as well as the readout of the front end electronics (FEE) and the spectrometer can be temperature dependent [1], there can be a detrimental effect on predictive model performance. A past reference measurement may no longer be representative current condition, thus guidelines on frequency of reference measurements need to be established to ensure and maintain optimal system performance. The objective of the current study was to model the spectral variation of quartz halogen lamps with respect to elapsed time from power up and variation in environmental temperature, and the variation in readout in the FEE and spectrometer with temperature, to aid in the development of guidelines to enhance predictive model performance.