We discuss the need for a new calibration of the mixing-length parameter owing to the inclusion of updated opacity in the computation of low-mass stellar models. When the contribution to the opacity of molecules is properly taken into account, a significantly larger value of the mixing length is required in order to reproduce the observed radii of the stars. In addition, at variance with previous claims, we find that different stellar models require different values of l/H-p. In particular we report some evidences of a correlation between mixing length and metallicity.