Thermodynamic modelling is a powerful tool in alloy development, and the CALPHAD method has proved to be successful when applied to complex materials with many components and phases, where limited experimental information is available and experimental determinations are time consuming. However, when extending to composition or temperature ranges that have not previously been studied, validation by experiments is needed. A few examples of experimental validation of thermodynamic databases for multicomponent steel and cemented carbides alloy systems are given. The choice of experiments and the interpretation of the results are discussed. The discrepancies between calculations and experiments that were found could be explained by inconsistencies in binary and ternary systems. The problems were due to lack of experimental information, or a need for better descriptions of metastable phases. To increase the ranges where reliable extrapolations can be made, critical experiments were designed to determine interactions in ternary systems. The methods to select critical experiments, and the resulting improvement of the thermodynamic description is discussed.