Purine compounds such as caffeine and theobromine are involved in cocoa flavour development. Measurement of these compounds is used to determine the percentage of cocoa mass in chocolate. Additionally the relative level of these two compounds may be associated with the genetic and/or geographical origin of the cocoa. The present work was undertaken to assess the potential of NIRS, as a rapid and non-destructive method for measuring these compounds. The study was carried out on 284 cocoa samples over 5 production years (1999 to 2003). The samples came from Ivory Coast, Venezuela and Trinidad, and belonged to genotypes of the Forastero, Criollo and Trinitario genetic groups. The sampling protocol ensured good representation of the spectral and chemical variability of fermented and dried cocoa. Spectra acquisition was perform on 3 g of shelled ground and sieved beans in diffuse reflectance using a FOSS Nirsystem 6500. Partial Least Square models (PLS) were used to establish quantitative relations between NIR spectral data and caffeine and theobromine contents. Models were developed which fitted the data with coefficients of determination of 0.95 for caffeine content and 0.89 for theobromine content. The SECV errors were 0.04% for caffeine and 0.07% for theobromine. These values are close to the repeatability of the reference method. The ratios of SD to SECV were 3.5 and 2.6 for caffeine and theobromine respectively. The theobromine to caffeine ratio separated Ivorian Forastero cocoa from the Trinitario cocoa from Trinidad and the Criollo cocoa from Venezuela. Given the performance of the equations developed to represent the range of samples over spectral variability, NIRS can be considered to provide a routine analysis method for purine compounds. Whilst this study did not lead to definitive conclusions on discrimination between genetic groups according to their purine compositions, it did confirm earlier observations. With a tool such as NIRS that enables determinatio