In this paper, we proposed a methodology to estimate cognitive function for action decisions in children by evaluating their behavior using a computational model for action decisions. First, we recorded the behavior of children while they were playing an immediate behavior-decision TV-game of “oni-gokko&rdquo. Then, we compared the recorded behavior with that of the computational model and adjusted its parameters so that the model behavior resembled that of the children. After adjusting, the model successfully reproduced more than half of the children's behavior. By observing the tendency of the model's parameter change from three to five years old, we found that the children's internal process changed: they changed their way of attention allocation to outer world objects and started to decide their action based on the objects in the wider area with deeper information processing. The results indicate that the children's action decision process, which is difficult to evaluate from merely the behavior recording, can be estimated more precisely by a computational model of their cognitive process.