Standard meta-analysis (Hunter & Schmidt, 1990a) assumes unidirectional (i.
e., all positive or all negative) predictor-criterion relations. We challen
ge this assumption in the context of personality-job performance linkages b
ased on several limes of evidence supporting the coexistence of true positi
ve and true negative correlations involving the same trait. Subjecting such
bidirectional relations to standard meta-analysis will underestimate effec
t sizes to an unknown degree owing to cancellation of positive and negative
values. A modification of standard procedures is proposed that accounts fo
r the possibility of bidirectionality. It employs successive iteration of a
n initial estimate of the absolute correlation, accounting for sample sizes
and the sampling distribution of the correlation. Tests of the procedure u
sing hypothetical distributions show it works as expected. Application to p
reviously documented personality-job performance relations (Tett, Jackson,
& Rothstein, 1991) yields interpretable effect sizes substantially stronger
than previously reported estimates. Implications for interpreting prior me
ta-analytic findings in this area are discussed.