We propose a modification to the popular 6-31G(*) basis set, which has recently been extended to cover first-row transition metals [Rassolov , J. Chem. Phys. 109, 1223 (1998)]. As demonstrated by a number of calculations, the existing basis performs poorly for many transition metals, particularly those toward the end of the series (Co, Ni, and especially Cu). The reason for this lies primarily with the 3D shell, which lacks a sufficiently diffuse exponent. A reoptimization of the D-shell exponents and coefficients by a two-step procedure, keeping the rest of the basis unchanged, corrects the problem, giving a basis set that performs uniformly well across the entire first-row transition metal series from scandium to copper. (C) 2003 American Institute of Physics.