In this article, one of the more important dilemmas in molecular physics is considered: given a matrix of the nonadiabatic coupling terms of any desired dimension, what is the minimal sub-Hilbert space for which diabatization is still valid. This problem was addressed by one of us before (Baer, M. Chem. Phys. Lett. 2000, 329, 450), but it was recently established that the suggested criteria therein lead to subspaces that are too large to be of any use. In this article, we discuss the conditions that have to be satisfied to reach the minimal subspace. We have found that these conditions are related to the spatial distribution of the various nonadiabatic coupling terms. Thus, if nonadiabatic coupling terms for the relevant states overlap only slightly in configuration space, the required size of the subspace for diabatization can be reduced significantly. As an example, we consider the C2H molecule.