This paper aims to establish a number of conditions under which generalization from single case studies is possible. The core of the argument is that these conditions are determined significantly by the antecedent theories that researchers bring to any case study. These theories, used to describe phenomena, employ general terms that go well beyond the particulars observed. Four areas of relevant background theory are examined that are of particular relevance to educational researchers. First, many people studied in educational research occupy roles which are specified to apply more widely by legislation, regulation, or job description. Second, a number of problems studied are very well defined admitting of solutions that would be acceptable in nearly any context in which the problems arise. Third, many studies of people's perceptions trade on common understandings of key terms in the questions asked, in short, commonsense theories embedded in language. However, these common understanding often embody enough common theory to admit of similar, or even identical, answers. Finally, knowledge of relevant substantive theories, for example, accounts of learning, child development, or policy, can be used to generalize beyond cases. The form of inference used to make generalizations, it is proposed, is not induction or deduction, but rather abduction, or inference to the best explanation. Finally, the paper deals with an important problem that is thought to occur in bringing theory to case study research: the problem of confirmation bias. It is proposed that since theory is ubiquitous, the key move is to conduct research with an eye to improving antecedent theory in the course of the research rather than to limit the amount of such theory. The paper concludes with a discussion of a number of epistemic strategies for doing this that employ an epistemology that coheres with the logic of justification by inference to the best explanation, notably a form of coherence justification.