Several innovative software applications such as those required by ambient intelligence, the semantic grid, e-commerce and e-marketing, can be viewed as open societies of heterogeneous and self-interested agents in which social order is achieved through norms. For agents to participate in these kinds of societies, it is enough that they are able to represent and fulfill norms, and to recognise the authority of certain agents. However, to voluntarily be part of a society or to voluntarily leave it, other characteristics of agents are needed. To find these characteristics we observe that on the one hand, autonomous agents have their own goals and, sometimes, they act on behalf of others whose goals must be satisfied. On the other, we observe that by being members, agents must comply with some norms that can be in clear conflict with their goals. Consequently, agents must evaluate the positive or negative effects of norms on their goals before making a decision concerning their social behaviour. Providing a model of autonomous agents that undertake this kind of norm reasoning is the aim of this paper.