We critically examine two nominally equivalent approaches for treating a random alloy: (1) using one very large supercell as a direct simulation of the alloy, and (2) performing configuration averaging over many smaller supercells; and the common practice using a virtual-crystal as the reference for analyzing the alloy band structure and discussing the electronic transport in the alloy. Specifically, (1) we show that in practice the size of the "very large" supercell depends on the particular property of interest, and the ideal of the configuration averaging is only useful for certain properties; (2) we examine the assumed equivalency by comparing the results of the two approaches in bandgap energy, energy fluctuation, and inter-valley and intra-valley scatterings, and conclude that the two approaches often lead to non-equivalent physics; (3) by using a generalized moment method that is capable of computing the global electronic structure of a sufficiently large supercell (e.g., ~ 260,000 atoms), we are able to obtain the intrinsic broadening of a Γ-like electron state, caused by the "inelastic" intra-valley scattering, in a direct bandgap semiconductor alloy; (4) we demonstrate an efficient way to construct the effective dispersion curves of the alloy, with high accuracy for calculating effective masses and examining anisotropy and nonparabolicity of the dispersion curve; and (5) finally, we discuss the limitation of using virtual crystal approximation as the reference for evaluating the alloy scattering and studying the transport property.2