The processing of stationary random sequences under nonparametric uncertainty is given by a filtering problem when the signal distribution is unknown. A useful signal (S n ) n≽1 is assumed to be Markovian. This assumption allows us to estimate the unknown (S n ) using only an observable random sequence (X n ) n≽1 .The equation of optimal filtering of such a signal has been obtained by A.V. Dobrovidov. Our result states that when the unobservable Markov sequence is defined by a linear equation with Gaussian noise, the equation of optimal filtering coincides with both the classical Kalman filter and the conditional expectation defined by the theorem on normal correlation.