Consumers shopping at a downtown commercial area usually visit several places and shops while changing their purpose on the way and deciding where to purchase what commodity. Such behaviors, known as “kaiyu” in Japanese, are called “shop-around behavior”. Its precise definition is an evolution of consumers' triple decisions about destination, aim, and expenditure on their way to move around a downtown. While analyzing and predicting such consumers' behaviors is quite important for various objectives, it has been quite a difficult task because one must take into account not only the changes of above consumers' triple decisions at once on the one hand but also the complex factors affecting those decisions like consumers' profiles such as their age, sex, budget, shopping attitude etc., as well as the characteristics of shopping trip such as having or not having accompanying person, its arrival time, weekday or holiday of its occurrence, etc., at the same time on the other. In fact, most of all previous studies could have been paying only attention to the changes of visited places. Thus the problem of how to deal with simultaneous changes of triple decisions has long been left open. This paper first addresses this problem. To tackle the problem of modeling complex shop-around behaviors, we employ Bayesian Networks (BN) method, which is a recently developed new statistical method suitable for representing complex phenomena involving many influencing factors and uncertainties, i.e., random variables. The distinct character of BN is a directed acyclic graph (DAG) representation, which is composed of nodes of random variables and directed arcs expressing dependence relations among nodes. The purpose of this paper is twofold: First, we apply the structural leaning algorithm of Bayesian Networks to the actual data of consumers' shop-around behaviors obtained at Downtown of Fukuoka City, Japan to extract from the real data a hidden DAG structure that represents a relational structure among shops formed by consumers' shop-around behaviors. Second, by constructing a hypothetical DAG structure by hand for consumers' shop-around behavior, we apply the parameter learning algorithm to learn from the same real data the conditional probabilities for the nodes in the DAG in order to make probabilistic inference to extract the significant information from the real data. From the structural learning, we have extracted two DAG structures for both genders to find the fact that female visitors are more likely to seek variety in their shop-around. Based on the result of learning the conditional probabilities, we have provided two cases of inferences. One case has shown us the most utilized shop for waiting to meet but revealed the fact that visitors who use the shop as meeting are not likely to visit the same shop as other sopping purposes than meeting, which had been unknown to us. With these results, we have shown that the efficacy and applicability of the BN method for modeling complex consumers' shop-around behaviors.JEL Classification: C11, R11, R23