Ellsberg's two-color problem, known as an example of ambiguity aversion, has generated a great deal of interest among many researchers. However, an unsettled question remains regarding the conditions that lead to heightened or diminished ambiguity aversion. We conducted a series of experiments, which required participants to choose between a clear and vague bet. Results showed that participants did not always prefer the clear bet, and the ratios of those who indicated ambiguity aversion varied depending on the types of ambiguity. Furthermore, ambiguity aversion consistently decreased when participants were allowed to choose the ambiguity task they would perform. These results were interpreted in terms of the pattern of second order probability distributions and illusion of control.