Abstract: Multilayer perceptron neural network are introduce as a new method for dat assimilation and is used in DYNAMO meteorological model. The approaches emulates the Kalmar Filter data assimilation method avoiding recalculation of the gain matrix at each instant of assimilation. A new procedure for training the networks is also presented, based on a modification in the backpropagation algorithm. An Adaptive Extended Kalman Filter was used to provide examples for network training.