The aim of the article is to show a stochastic approach for both modelling and optimizing the statistical agent belief in a probability model.

Two networks are defined: a decision network $\mathfrak{D}$ of the agent belief state and a utility network $\mathfrak{U}$, presenting the utility structure of the agent belief problem.

The agent belief is presented via the following three items ($\mathfrak{B},\mathfrak{D},\mathfrak{U}$), where $\mathfrak{B}$ is a Bayesian network, presenting the probability structure of the agent belief problem.

Two propagation algorithms in $\mathfrak{D}$ and in $\mathfrak{U}$ are also presented.