The mid-infrared spectroscopy (MIRS) was investigated as a tool to improve the quality of tomato products considering its implementation at different steps along the processing chain. Models have been developed using partial least square (PLS) regression to predict the quality of raw and processed tomatoes. A relevant method (Multi-year Combining models) consisting in adding early-season tomatoes data within models developed using data of previous years, was shown as the most efficient and adapted to realistic industry conditions. MIRS predicted, in external validation, soluble solids content (R2 0.95), titratable acidity (R2 0.88) and dry matter content (R2 0.81) with a high accuracy of 0.1°Brix, 2.8 mmol H+/kg and 0.4% respectively. Secondly, MIRS was used to classify tomato products depending on processing methods (hot- or cold-break) or varieties using factorial discriminant analysis (FDA) based only on spectral data. MIRS was assessed as an efficient tool to classify processed tomato purees according to process, year and variety, more accurately than the classification obtained with the reference data. A possible implementation of MIRS was suggested at three strategic steps along the processing chain to i) characterize the incoming raw material, ii) monitor the matrix changes during processing and iii) control the final products.