We have applied the augmented plane-wave plus local-orbitals method to investigate the charge density of Cu2O by means of Bader's topological analysis and electric-field gradients (EFGs). It is rather clear that a simple Cu+-O2- model is inadequate for the explanation of bonding properties of Cu2O. Appearance of s-d hybridization in this system has been pointed out already in the literature. However, the amount of charge transferred from d to s Cu orbitals is overestimated by both the local-density approximation (LDA) and the generalized gradient approximation. As a result, the calculated EFG is underestimated by about 50% compared to experiment. Also the topological analysis of calculated densities suggests that density-functional theory overestimates the covalency of the Cu-O bond. In order to demonstrate the role of s-d hybridization as the main reason for the mentioned discrepancies, we have utilized artificially modified basis sets. Removing part of the 4s Cu character from the wave function increases the absolute EFGs and decreases the calculated bulk modulus, both resulting in better agreement with experiment. This is a result of an increased Cu d occupation and consequently a decreased asphericity of the d charge distribution. A more physical description of the localized nature of the 3d orbitals should be given by the LDA+U correction. We find that LDA+U "self-interaction corrected" (SIC) by Anisimov [V.I. Anisimov, I.V. Solovyev, M.A. Korotin, M.T. Czyzyk, and G.A. Sawatzky, Phys. Rev. B 48, 16 929 (1993)] leads to EFGs in good agreement with experiment while LDA+U "around mean field" Cyzyk and Sawatzky [Phys. Rev. B 49, 14 211 (1994)] even worsens the results. This is to be expected according to the analysis given above, which indicates that the on-site correlations are important but are implemented only in the LDA+U (SIC) method.