The ability of different quantum chemical methods to predict experimental electronic circular dichroism (CD) spectra is critically evaluated. Two single-reference, time-dependent approaches based either on density functional theory (TDDFT) or a simplified coupled-cluster expansion (CC2) and two multireference methods (MRMP2 and DFT/MRCI) are considered. The methods are applied to a test suite of seven molecules including a wide range of difficult chromophores ("real-life" examples) and to three model systems-H2S2, twisted ethylene, and dimethyloxirane-where accurate ab initio MRCI reference data are used for comparison. To investigate the effect of "exact" exchange mixing systematically, the TDDFT calculations were carried out with the BP86, B3-LYP, and BH-LYP functionals. The time-dependent Hartree-Fock (TDHF) method was included as an "upper limit" for the HF-exchange part in the functional. In general, it is found that the accuracy of most of the simulated spectra (except those from TDHF) is good enough to assign absolute configurations of chiral molecules with very high certainty. However, the description of weakly disturbed, inherently achiral chromophores and systems with Rydberg-valence mixing turned out to be rather difficult. Furthermore, none of the methods perform reliably for all of the molecules in the test suite, and in particular, the TDDFT results are very sensitive to the functional used. The best overall performance is achieved with the DFT/MRCI and CC2 methods. The TDDFT method should be used carefully, especially for systems with important diffuse or charge-transfer states. Out of the three functionals tested, B3-LYP seems to perform best. In practice, we highly recommend the simultaneous application of different complementary single and multireference methods, which significantly increases the reliability of theoretical predictions.