Low-frequency molecular rotations recorded in a region of 0–150 cm^–1 with a high “signal/noise” ratio in the ultrafast optical Kerr effect (OKE), which is a third-order nonlinear optical response upon nonresonant excitation of a liquid by femtosecond laser pulses, are analyzed in detail. It is shown that the reduced Raman spectral density (RSD) derived from experimental data using the well-known deconvolution procedure is ambiguous due to the problem of separating the orientational and librational contributions to the total OKE signal. This fact significantly limits the reliability of information derived about molecular motions in a liquid. For the example of an ultrafast OKE in benzonitrile, it is shown that, in the range of 0–300 fs, rotational responses cannot be considered independent, and the application of a number of additional criteria resulting from the assumption of their correlation allows this ambiguity in deriving the RSD function from experimental data to be removed.