Any serious empirical study of factor substitutability has to allow the data to display complementarity as well as substitutability. The standard approach reflecting this idea is a translog specification - this is also the approach used by numerous studies analyzing the relative capital-skill complementarity hypothesis formulated by GRILICHES (1969). According to this hypothesis, the degree of substitutability between skilled labor and capital is lower than that for unskilled labor and capital. Yet, the results of empirical studies investigating this hypothesis are controversial. This paper offers a straightforward explanation: Using a translog approach reduces the issue of factor substitutability or complementarity to a question of cost shares. Our review of translog studies mentioned in HAMERMESH’s (1993) summary on the demand for heterogeneous labor demonstrates that this argument is empirically relevant - all these studies can be reconciled with each other on the basis of the cost-share argument.