Using data observed on crop status to inverse a crop model is an interesting way of estimating the soil parameters, which are rather difficult to determine. Nevertheless, the results of parameter estimation depend on the observation set and the results of the predictions made with the model are also affected. The goal of this study is to assess the quality of soil parameters estimation and of predictions of selected variables of interest for various observation sets. Several synthetic and real observation sets acquired in different conditions (winter wheat and sugar beet crops grown in different weather, soils and cropping conditions) were used to estimate the values of the soil parameters. These estimates were then reused in the model to predict agro-environmental variables of interest for wheat and sugar beet crops. Parameters were estimated using the Importance Sampling method. The results show that the estimate of parameters related to soil water content and soil depth can be significantly improved as compared to prior information. Concerning the prediction of variables of interest, the best results are obtained for wheat yield. Improvement in both parameter estimation and variable prediction depends on the observation dataset: it is greater for conditions were water stress effects are important, that is for shallow soils, dry weathers and when the observations are made on sugar beet crops