Vis-NIRS, MIRS, and a combination of both coupled with PLS and machine learning were applied to i) trace the composed proportions of different apple varieties in formulated purees and ii) predict the quality characteristics of formulated purees from spectral information of initial puree cultivars. The PLS models could estimate proportions of each apple cultivar in puree mixtures using MIR spectra (RMSEP<8.1%, RPD> 3.6), especially for Granny Smith (RMSEP = 2.7%, RPD = 11.4). The concentration profiles from multivariate curve resolutionalternative least squares (MCR-ALS) made possible to reconstruct spectra of formulated purees. MIRS technique was evidenced to predict the final puree quality, such as viscosity (RPD>4.0), contents of soluble solids (RPD = 4.1), malic acid (RPD = 4.7) and glucose (RPD = 4.3), based only on the spectral data of composed puree cultivars. Infrared technique should be a powerful tool for puree traceability, even for multicriteria optimization of final products from the characteristics of composed puree cultivars before formulation.