Near infrared (NIR) diffuse reflectance was used to predict the contents of taste-related compounds of tomato. Models were obtained for several varietal types including processing tomato, cherry and cocktail tomato, mid-sized tomato and tomato landraces, with a wide range of varieties. Good performance was obtained for the prediction of soluble solids, sugars and acids, considering a non-destructive methodology applied to fruits with different internal structure. Specific models averaged RMSEP (%mean) values lower than 6.1% for SSC, 13.3% for fructose, 14.1% for glucose, 12.7% for citric acid, 13.8% for malic acid and 21.9% for glutamic acid. The performance was dependent on varietal type. General models with a higher number of samples and variation did not improve the performance of specific models. The models obtained, either specific or general, couldn't be extrapolated to external assays and an internal calibration would be required for each assay in order to provide a reliable performance.