In this paper, we propose a method of coalition formation for assigning tasks to appropriate aGents to improve the effciency of multia-Gent systems. To form a coalition, we introduce subjective information to aGents, which are the internal information of the aGents. The subjective information reflect the aGents' cooperative behavior of the past. Next, we introduce loose coalition, a concept of a coalition of aGents based on the subjective information. Using the aGents' sense of values defined by their subjective information, each aGent can give priority to the loose coaliti-ons to ask for the working status or to assign tasks. Thus loose coalitions with higher priority will be better cooperating candidates. Furthermore, loose coalitions enable aGents to collect information (e.g. busyness of loose coalitions) for task assignment effciently. Therefore, the aGents on the system can decide its behavior properly, depending on the current status of the system, and thus the effciency of the system can be im-proved. Then, we observe dynamic properties of system under several settings of aGents to derive a guideline for designing effective multiaGent systems based on loose coalitions.