Using as benchmarks a series of increasingly large hydrated uracil clusters, we examine the nature and extent of charge-transfer (CT) contamination in condensed-phase, time-dependent density-functional theory. These calculations are plagued by a large number of spurious CT excitations at energies comparable to (and sometimes below) the valence excitation energies, even when hybrid density functionals are used. Spurious states below the first nπ* and ππ* states of uracil are observed in clusters as small as uracil-(H 2 O) 4 . Reasonable electronic absorption spectra can still be obtained, upon configurational averaging, despite pervasive CT contamination, but the spurious states add significantly to the cost of the calculations and severely complicate attempts to locate optically dark nπ* states. The extent of CT contamination is reduced substantially by introducing an electrostatic (point charge) description of an extended solvent network, even in cases where the region of solvent described by density functional theory is large (>120 atoms). Alternatively, CT contamination may be reduced by eliminating certain excitation amplitudes from the linear response equations, with minimal loss of accuracy (<0.1 eV) in the valence excitation energies.