Near Infrared Spectroscopy (NIRS) is a rapid and non-destructive technique, which does not use chemicals but can simultaneously determine numerous constituents in root crops. Spectrophotometers are simple to manipulate and the simple spectral collection reduces the risks of operator errors. However, NIRS is not a stand-alone technology and its performance depends upon the relationships between the spectral information and an accurate chemical method to measure the constituents of interest. The major constraint is, therefore, the need to build a reliable equation of prediction, also called a calibration model. This paper reviews the practical steps involved in the development of a model, including: 1. the collection of highly variable samples for calibration and validation sets; 2. the production of chemical data for these samples; 3. the collection of spectra on samples from the calibration and validation sets; 4. the development of the multivariate model; 5. its validation using descriptive statistics and 6. the monitoring of the model performance. The paper also reviews recent studies conducted on six root crops, including four tropical species (cassava, sweet potato, yam and taro) and two temperate (potato and sugar beet). Their results and the perspectives for further studies are discussed. (Résumé d'auteur)