The evolution of microstructures of steels is controlled by appropriate thermo-mechanical processing. In this sense, cooling after austenitization of steels is a crucial task and is thus investigated experimentally by dilatometry, metallographic investigations and X-ray diffraction. The X-ray diffractograms are analyzed by the classical Rietveld method combined with the Double-Voigt approach. Whereas dilatometry, metallographic investigations and hardness measurements help to distinguish between martensite, ferrite, bainite and pearlite, evaluating the X-ray diffractograms allows quantifying the phase fraction of austenite. The experimental investigations are evaluated by machine learning algorithms with the potential to relate microstructures to material properties. The classification of martensite, ferrite, bainite and pearlite is supported by an unsupervised machine learning algorithm (here hierarchical clustering) on the basis of measured X-ray diffractograms in a first step. In the second step material properties like hardness data and phase fractions are linked with data deduced from X-ray measurements by partial least squares regression. Cross validation is used to determine the quality of this supervised machine learning approach and indicates possible overfitting. Thereby, the possibilities and also the limits of the used numerical tools are discussed. It is likely that various applications in quality control during steel processing are supported in future by data evaluation tools similar to those presented.