Aim: The aim of the present study was to use Fourier transform infrared (FT-IR) spectroscopy with chemometrics to develop partial least squares (PLS) models to predict the concentrations of various anthocyanins during red wine fermentation. Methods and results: Must and wine samples were collected during fermentation. To maximize diversity, 12 different fermentations, of two different vintages and two different varieties, were followed. The anthocyanin composition of the samples was characterized by using different reference analyses described in the literature: the concentration of free anthocyanins was determined by bisulphite bleaching, the concentration of molecular anthocyanins was determined by high-performance liquid chromatography with ultraviolet-visible detection, and the ratio of monomeric anthocyanins to polymeric anthocyanins was determined using the Adams-Harbertson assay. Finally, the data were analysed statistically by PLS regression to quantify laboratory-determined anthocyanin from FT-IR spectra. The correlations obtained showed good results for a large percentage of parameters studied, with the determination coefficient (R-2) for both calibration and cross-validation exceeding 0.8. The models for molecular anthocyanins appeared to overestimate their concentration, owing to intercorrelation with other parameters. Comparison of the data for each vintage indicated no apparent matrix effect per year, and data for other vintages should be compared to validate this hypothesis. The best models were those for monomeric or polymeric pigments and free anthocyanins. Conclusions: By using FT-IR spectroscopy coupled with chemometrics, it is possible to create predictive models to estimate concentrations of anthocyanins and changes in global anthocyanin parameters during winemaking. Significance and impact of the study: These results improve our understanding of anthocyanin prediction using FT-IR spectroscopy with chemometrics and pave the way for its use as a process control tool for the winemaker. They also highlight the propensity of chemometrics to overestimate certain predicted values when close parameters intercorrelate.