The accuracy and convergence of a series of commonly used split-valence electronic basis sets are systematically investigated for the anharmonic molecular vibrational spectroscopic calculations using the second-order perturbative corrected vibrational self-consistent field theory. A series of 18 split-valence basis sets with increasing flexibility is assessed in conjunction with MP2 and density functional theory (DFT)-based dispersion-corrected B3LYP potentials and applied to a set of molecular systems with different electronic and vibrational characteristics. The computed fundamental transitions and intensities are compared with the experimental values to assess the accuracy of different basis sets. Comparisons with high levels of Dunning basis sets, aug-cc-pVQZ along with def2-TZVPP and aug-pc-2 type basis sets, are also tested to check the convergence. A full statistical error analysis has been performed, assessing the success and failure of each basis set in terms of both accuracy and precision. The general statement "the bigger the basis, the better the results" is not strictly followed for bigger basis sets that are nearly converged to the adiabatic ab initio limit. It is found that the basis set hierarchy holds good for minimal basis sets up to polarized double- basis sets. Beyond that, the improvements in comparison to experimental results or with very high end aug-cc-pVQZ and others are found rather slow. Finally, a prescription regarding the choice of a basis set for a suitable balance between accuracy and the computational time is given, which can be further used to investigate specific normal modes and large molecules as a blackbox.