The two standard steps for estimating PERT times are: Step 1, estimate
a, m, and b; and Step 2, use the 'classical' formulae mu = (a + 4m b)/6 and sigma = (b - a)/6. We review the shortcomings of the textbook
definitions of a, m and b; we also review the inconsistency of Step 1
with the literature on probability elicitation. A 5- or 7-fractile al
ternative is then proposed and justified for Step 1. Next, we develop
simple but very accurate formulae for computing mu and sigma with the
fractiles estimated in our Step 1. For contrast, we also show that the
classical PERT formulae are very inaccurate, even for the very restri
cted subset of beta distributions for which the formulae are supposedl
y applicable. Our overall purpose is to combine earlier findings with
some new results to argue that: (i) the classical PERT formulae are bo
th illogical and inaccurate, so we should not continue to teach and us
e them; and (ii) simple and more logical alternatives are available.