Abstract: In the case of atmospheric continuous data assimilation there are many deterministic and probabilistic methods. A new approach based on neural networks is proposed. The new aproach adopted requires training of a multilayered perceptron to emulate a chosen data assimlation method. Here, Kalman filter data assimilation methods were used to generate training examples for the networks. Tests are performed using the Henon mapping and Lorenz evolution system. In the case of Henon system an Adaptive Extended Kalman Filter is used to provide examples for network training. In the case of Lorenz system the Extended Kalman Filter is used for network training. The preliminary results obtained are promising.