When parametric nonlinear processes are employed in the cause of efficient optical frequency conversion, the media involved are generally subjected to substantially off-resonant input radiation. As such, it is usually only electronic ground states of the conversion material that are significantly populated; higher levels are engaged only in the capacity of virtual states, and it is frequently assumed that just one such state dominates in determining the response. Calculating the nonlinear optical susceptibilities of molecules on this basis, excluding all but the ground and one excited state in a sum-over-states formulation, signifies the adoption of a two-level model, a technique that is widely deployed in the calculation and analysis of nonlinear optical properties. The two-level model offers tractable and physically simple representations of molecular response, including wavelength dependence; it is also the origin of the widely applied 'push-pull' approach to designing optically nonlinear chromophores. By contrast, direct ab initio calculations of optical susceptibility are commonly frustrated by a complete failure to determine such dispersion features. However, caution is required; the two-level model can deliver potentially misleading results if it is applied without regard to the criteria for its validity, especially when molecular excited states are significantly populated. On the basis of a precise, quantum electrodynamical basis for the theory, we explore in detail why there are grounds for questioning the general validity of two-level calculations in nonlinear optics; we assess the criteria for high frequency conversion efficiency and provide a new graphical method to assist in determining the applicability of a two-level model for hyperpolarizability calculations. Lastly, this paper also explores the applicability and detailed conditions for the two-level model for electronically excited molecules, identifying problematic results and providing tractable methods for improving the accuracy of calculations on real molecule-photon interactions.