In the studies of deductive reasoning, the experiments have always been conducted with tasks with just one logically correct answer. In this paper, three experiments examined how the reasoning process with underspecified tasks can be explained. Underspecified tasks are tasks with an arbitrary correct answer or with no correct answer at all. Experiment 1a and 1b showed that participants tend to choose the option that is more difficult to draw a conclusion than the other. Furthermore, the factors leading to such selection bias included the differences in the number of logical symbols contained in the reasoning process between options and the reductio ad absurdum process. Protocol analysis in Experiment 2 revealed that reductio ad absurdum has more impact on selection bias.