Abstract: The numerical weather prediction is a process involving many steps: data collection, data quality control, objective analysis, initialization, and numerical integration of atmospheric dinamic system. However, during this cycle other data are available. In order to impruve the prediction this new data should be somehow used in the integration. This procedure is called meteorological data assimilation. Many approaches have been suggested to do it. The Kalman filter is one of the techniques proposed. Classically, the Kalman filter uses a priori information on both modelling and measuring errors. In the end of the 60's an adaptative version of the filter appeared, which drops the need for a priori knowledge of the modelling error (an adaptative strategy). For the meteorological assimilation process an adaptative version was used by Burg et al. Here is investigated a simpler version proposed by Kuga and Rios Neto in the assimilation process. This version is applied to Lorenz's system in chaotic regime and to Dynamo meteorological model, which shows a good performance in our tests.