One of the coefficients that parametrizes the chiral Lagrangian of QCD is expressible in terms of a sum rule for which the integrand is known from experiment. We confront two model calculations of this coefficient with the behavior of the data. The model related to vector dominance reproduces the sumrule behavior very well, whereas a one-loop chiral quark model calculation fails badly. The sum rule allows one to identify those ingredients which are needed to make the quark method more realistic.