This article discusses Parikh's axiom of relevance in belief revision, and recalls some results from Kourousias and Makinson (2007, J. Symbolic Logic, 72, 994-1002) in this context. The crucial distinction is emphasized between the uniqueness of the finest splitting of K and the fact that K has several normal forms associated with that finest splitting. The main new result of this article is a new proof for the theorem that the set of prime implicates of K is a normal form for the finest splitting of K. It is explained how this proof avoids a mistake in an earlier proof from Wu and Zhang (2010, Knowledge-Based Syst., 23, 70-76). As a corollary, relevance can be redefined without reference to the finest splitting, using the notion of path-relevance from Makinson (2009, J. Appl. Logic, 7, 377-387). Finally, a weak yet sufficient condition for irrelevance is presented.