In conventional NMR spectroscopy, the data are recorded independently either in terms of longitudinal, T1, or transversal, T2, relaxation times. However, interpretation of these parameters could be very tedious in the case of molecule mixtures or in heterogeneous and porous systems. In order to understand and interpret the T1 and T2 distributions measured in such systems, T1-T2 NMR spectroscopy can be used [1, 2]. This approach is based on a joint observation with respect to the two relaxation parameters and a data processing based on a numerical 2D-Laplace inversion. It has proved to provide more robust results than one-dimensional measurements since 2D T1-T2 spectra reveal any correlation between T1 and T2 relaxation times, and can help in the interpretation of their variations under a chemical or physical modification of the sample [3, 4]. We performed T1-T2 NMR spectroscopy to monitor the starch transition processes occurring upon hydration and temperature change of starchy samples. Data were processed using an efficient method, based on maximum entropy regularization and truncated Newton optimization [5]. We will show that 2D T1-T2 correlations permit a better understanding of the water distribution during the thermal processing of starch in model systems and dough.