This paper presents an abstract general model for representing the belief systems of resource-bounded reasoning agents. The intuition which underlies this new model is that it is possible to capture the key properties of many different types of belief system in a structure called a belief extension relation. The paper shows how such a relation may be derived for any system that satisfies some basic properties. The resulting formalism is simple, and yet suf-ficiently rich that it generalises many other frameworks for representing belief. A logic is defined, using the new model to give a semantics to belief modalities. The properties of the model and logic are discussed in detail. The paper closes with a discussion and remarks on future work.