In this paper the observation process of stochastic discrete-time nonlinear system is analyzed. The system to be observed is assumed to be uncertain, but fulfilling the global "quasi-Lipschitz" condition and is subjected to stochastic input and output disturbances of a white noise type. The combination of a traditional Luenberger residual term with a discontinuous one is considered. The designing of the best observer gain matrices is realized by using the Robust Attractive Ellipsoid Method for the analysis of the averaged observation error. The construction of this attractive ellipsoid is based on the numerical solution of some matrix optimization problem under specific constraints of Bilinear and Linear Matrix Inequalities (BMI's and LMI's) type applied to improve the attractiveness zone estimation. Two numerical xamples illustrate the effectiveness of the suggested approach. K E Y W O R D S attractive ellipsoid method, bilinear and linear matrix inequalities, discrete-time stochastic process, Quasi-Lipschitz conditions, sliding mode observer 1 INTRODUCTION The problem of state variables estimating of a dynamic system, given observations of the output variables, is of fundamental importance in control theory since many feedback control designs require the availability of the state of the controlled plant. 1.1 Complete information on dynamics and conditional probabilities 1.1.1 Linear estimations Considering the class of linear systems, there are two approaches available:-if the output variables can be measured exactly and if there are no stochastic disturbances acting on the system, then one can use a Luenberger observer to reconstruct the state 1-3