In most applications, the
probability distribution of random variables is unknown or if it is given, it would be too expensive
to consider the discrete
distribution with a huge possible realization or to handle the continuous distribution with numerical integration. It is common to choose a set of representative realizations with relatively small in number called scenario to present random events. Scenario can be a quartile of a known distribution or historical data, prediction of several trees or constructed using simulation. Each scenario is assigned to a
probability value to reflect the likelihood of the occurrence of a random event. For multi-stage model the information of scenario can be organized in a tree structure. In this paper we purpose an algorithm for generating efficiently tree decision of multi-stage stochastic programming problem. A heuristic method is used to generate discrete probability distribution specified by four first marginal moment and correlation.