This paper considers the objective function of the Additive (Pareto-Ko
opmans) Data Envelopment Analysis model and finds a flaw which is indu
ced by the manner in which units invariance is sought. This flaw is re
ctified, in the spirit of the original proposal, but the consequence i
s a non-linear objective function for the efficiency calculations. How
ever, the resulting non-linear programming problems solve readily with
existing NLP software.