Since the exact Expected Utility Functions (EUF) are not available in many circumstances, it is hard to make decision in environment characterized by incomplete information and uncertain decision results. Being aware of the defects of traditional decision analysis techniques, a new decision making method named Evolutionary Decision Making Based on Candidates Ranking is proposed. By selecting a set of indexes relevant to the expected utility of the candidates, the construction of decision rules can be reduced to finding the quantitative relationship between them. If all of the candidates are classified according to the indexes relevant to their expected utility, then Evolutionary Algorithms can be used to search for the expected utility ranking of the whole set of candidate classes, thus the optimal decision can be made based on the ranking. Some special considerations for Genetic Algorithms for ordering problem are also highlighted. The new method enjoys the advantages of weak dependence on expert knowledge, robustness in environment with random noise, no dependence on explicit EUF, effectively treating non-numerical, non-quantificational indexes and conflicts or correlation among indexes. The effectiveness of the proposed method is validated by its successful application in the controller design of certain simulated robot.