Abstract Some applications of the Wiener-Kolmogorov and Bayes approaches are discussed with regard to the construction of recursive equations for the best linear estimator of a stochastic parameter. Solutions are given for the prediction, filtering and smoothing cases when the parameter follows a vector Markov process and is observed with error. The Bayes solution is used to generate a canonical factorization of the autocovariance generating function of the observed series. Generalization of these techniques to other models is also discussed.