Knowledge-based systems (KBSs) are being applied in ever increasing nu
mbers. In parallel with the development of knowledge acquisition tools
is the demand for mechanisms to assure their quality, particularly in
safety critical applications. Quality assurance is achieved by checki
ng the contents of the KBS at various stages throughout its life cycle
. But how does testing for quality assurance aggravate the already wel
l-known knowledge acquisition bottleneck? The partial automation of ch
ecking and correcting the knowledge base (KB) is an obvious approach t
o reducing the bottleneck, but also a more routine treatment of checki
ng will provide improved facilities for quality assurance. In addition
to identifying the occurrence of faults, this paper suggests that res
ponding to faults identified by validation is both useful and importan
t. Therefore, refinement should be thought of as a companion to valida
tion.