This topic as been (too often) present in our discussions and deserves an entry on it's own. If you want, you can add your own views, and/or other citations, and/or contrary views ;-)
Kathleen M Eisenhardt & Melissa E Graebner (2007). Theory building from cases: opportunities and challenges. Academy of Management Journal, vol. 50(1), pp. 25-32:
"(...) frequent challenge to theory building from cases concerns case selection. Some readers make the faulty assumption that the cases should be representative of some population, as are data in large-scale hypothesis testing research. In other words, they ask, How can the theory generalize if the cases aren’t representative?(...) clarify that the purpose of the research is to develop theory, not to test it, and so theoretical (not random or stratified) sampling is appropriate. Theoretical sampling simply means that cases are selected because they are particularly suitable for illuminating and extending relationships and logic among constructs. (...) cases sampled for theoretical reasons, such as revelation of an unusual phenomenon, replication of findings from other cases, contrary replication, elimination of alternative explanations, and elaboration of the emergent theory."