Given a specification, a partition testing strategy is said to be safe for
a program if it is not less probable of detecting at least one failure than
random testing, and universally safe, if it is safe for every program. The
proportional sampling strategy (PSS) has previously been proved to be univ
ersally safe. This paper extends previous results by proving that PSS is ac
tually the only universally safe test allocation strategy for partition tes
ting, thereby settling an open problem in the search for universally safe s
trategies. Despite this, we show that a close approximation to PSS may stil
l be safe for most programs. We also discuss the practical implications of
this result. (C) 2000 Elsevier Science Inc. All rights reserved.