Statistical decision theory is a framework with a probabilistic foundation, which admits the random uncertainty about the real world and human thinking.In this paper, we propose a theoretical framework for the observational data oriented decision analysis under general uncertainty environments.We start with the three basic elements of an uncertainty decision problem, evolve naturally to the investigation on the unique characteristic and the fundamental roles of uncertainty distribution underlying the observational data in decision process, and finally end at the establishment of an uncertainty Bayesian data modeling under the Maximum Uncertainty Principle.