Large bodies of knowledge require careful consideration of the retrieval problem. If packets of knowledge can be identified that account for special cases in the application area, they might be left unaccessed when other knowledge is found relevant. The paper discusses principles for the organization of structured knowledge bases, aspects of granularity important for structuring knowledge, and the need for mechanisms to mediate changes among different levels of detail. Significant clusters of domain con-cept words, possibly identified by associanist networks, are considered promising candidates to guide context-driven knowledge retrieval with respect to specificity as well as to granularity.