The accuracy of scalar-relativistic energy-consistent small-core pseudopotentials has been tested in quantum Monte Carlo calculations for the first-row transition metals Sc, V, Ti, and Cr. We have calculated atomic ionization and excitation energies using variational and pure diffusion quantum Monte Carlo methods as well as coupled cluster theory with nearly complete basis sets up to g functions. On the basis of the results we are able to estimate the magnitude of various errors related to the use of pseudopotentials in quantum Monte Carlo calculations for transition metals. From a comparison of coupled cluster valence-only and all-electron results, respectively, of the experimental data we estimate the errors due to the semilocal pseudopotential to be at most 0.1 eV. The average error of pure diffusion quantum Monte Carlo results with respect to experimental data amounts to 0.2 eV and arises mainly from errors due to the semilocal pseudopotential, its localization and the fixed-node approximation used in the quantum Monte Carlo calculations. The errors appear to be closely related to the occupation of the 3d shell. (C) 1997 American Institute of Physics.