Recently, we reported that the basis set has a profound influence on the computed values for spin-state splittings [J. Phys. Chem. A 2008, 112, 6384]. In particular, small Gaussian-type orbital (GTO) basis sets were shown to be unreliable for the prediction of them. Here, we report simple modifications of the small Pop le-type Gaussian-type orbital basis sets (3-21G, 3-21G*, 6-31G, 6-31G*), which correct their faulty behavior for the spin-state energies. We have investigated the basis sets for a set of 13 first-row transition-metal complexes for which reliable reference data have been obtained at the OPBE/TZ2P(STO) level. For several systems, we have used single and double spin-contamination corrections to avoid ambiguity of the results because of spin contamination, that is, the energies and geometries were obtained for the pure spin states. The spin ground states as predicted by the spin-state-corrected GTO basis sets (s6-31G, s6-31G*) are in complete agreement with the reference Slater-type orbital (STO) data, while those of the original basis sets and a recent modification by Baker and Pulay (m6-31G*) are not for all cases. The spin-state-corrected GTO basis sets also improve upon the original and modified basis sets for the accuracy of geometry optimization, while the accuracy of the vibrational frequencies is as good or better. At a limited additional cost, one therefore obtains very reliable results for these important spin-state energies.