The agreement between the estimates and an independently calculated value of the activation energy, and the least squares criterion, have been used as alternative criteria for discriminating between rivalling kinetic models of the non-isothermal decomposition of a solid. It is shown that the two criteria are in good agreement when the data embrace a wide range of thermal curves with different heating rates.The conditions are discussed for determination of the most accurate and precise estimates of the activation energy and the other model parameters by the non-linear estimation approach.The discussion is supported by calculations performed on experimental results of dehydration of CAC204 ' H~O and Ba(IOz)~ 9 HzO.Determination of the activation energy and other kinetic parameters of thermal decomposition of solids from thermogravimetric experiments can be defined as a problem of model identification. Different methods of identification used by various authors [1, 2] require specific types of data, collected under special experimental conditions, e.g. differential or integral data from isothermal runs or from runs conducted at constant heating rate. As a rule, individual methods are valid only for selected forms of kinetic models.The development of various methods for identification of decomposition models stemmed from a desire to simplify laborious procedures for mathematical processing of experimental results. However, in the elaboration of these methods some assumptions were made that are difficult to fulfil in real experiments, e.g. linear temperature rise.Nowadays the advance in numerical, computer-oriented methods, such as the non-linear least squares, maximum likelihood, quasi-linearization and others [3][4][5], allows one to overcome the computational problems of identification. Effective optimization procedures are capable of handling models of any complexity. The problem is how to choose from among the various possible models. the one that would be applicable over the widest range of experimental conditions.The criteria for the planning of experiments have also undergone some changes. Instead of the experiments being planned to make the measurement data fulfil the assumptions of a simplified method for data treatment, the experiments are now designed to provide the maximum volume of information on individual parameters. It is important that studies be carried out under experimental conditions 1"