The objective of this paper is to investigate and compare two model-based methods for the soil moisture retrieval from SAR data. The overall accuracy of model-based methods in estimating geophysical parameters mostly depends both on the performances of the exploited direct model and on the intrinsic ambiguity of SAR data. The ambiguity is due to the complexity of the relationship between the geophysical parameters, such as soil moisture and soil roughness, and the backscattering values that makes 'ill posed' the inverse problem of the parameter retrieval. Moreover, the accuracy of soil moisture estimates depends also on the retrieval algorithm and on its robustness to the noise. In this study, the methods under investigation perform a probabilistic estimation of the parameters, finding solutions representative of an unknown distribution such as the mean or the most probable solutions. The model-based methods considered are a Neural Network algorithm, to explicit invert the direct model, and a Mixture Model algorithm, to approximate the parameter distribution function. The theoretical direct model adopted to tune the inversion algorithms is the Integral Equation Method (IEM) model. A comparison of the characteristics of the two algorithms is shown, and same evaluations of the accuracy in predicting soil moisture content from SAR data are performed. In particular, evaluations refers to ERS and ENVISAT ASAR data simulated by means of the IEM model.