The measured parameters of spin-orbit spectral splitting in HgCdTe-based MIS structures with positive and negative Kane gap E _g are compared with the parameters calculated using the three- and four-band Kane model. The disregard of the finite spin-orbit splitting A of the valence band in calculations leads to exaggerated values of Rashba splitting (especially for E _g < 0) even for small ratios |E _g |/Δ, although the subband parameters averaged over two spin branches of the spectrum in the two-, three-, and four-band Kane approximations for the same concentrations are practically identical. In the zero-gap HgCdTe, the measured as well as calculated values are noticeably higher, but the four-band approximation leads to values of splitting for both materials which are 20-40% lower than the experimental value. The inclusion of the interband interaction reduces these discrepancies, but does not eliminate them completely. It is shown that the approximations of the 2D spectrum with spin-orbit splitting linear in quasimomentum, which are conventionally used in the analysis, may lower the effective Rashba parameter by a factor of 2-4.