The emergence of nanotextures in photovoltaics has resulted in challenges associated with optical modelling. Whilst rigorous methods exist to accurately solve these textures, the computational effort required limits the scope of modeling applications. The effective medium approximation (EMA) is a potential alternative to provide rapid modeling results which can be easily integrated with ray tracing of large complex structures. However, the validity of this technique is strongly dependent on the size of features relative to the wavelength of interest, making the application of EMA ambiguous for many situations. This paper aims to address this issue by comparing the simulated results between EMA and finite element methods for three randomly distributed silicon textures with and without a dielectric layer. Criteria for which the EMA approach is valid are proposed and generalized using ratios between root-mean-square roughness, correlation length and incident wavelength, making these limits broadly applicable, beyond that of just the nanotexture under specific solar spectrum regimes. The results in this work apply to random, isotropic textures under normally incident light. Based on the proposed criteria, the validity of different optical simulation techniques for a set of industrial photovoltaic textures is discussed. This analysis reveals a region within which neither geometric optics nor EMA are adequate for calculating the reflectivity of a textured surface, and hence FDTD or other new approaches are required.