The two-fluid theory of binary mixtures postulates that the extensive thermodynamic properties of a binary mixture may be expressed by the contributions of two hypothetical fluids that mix ideally. This postulate, coupled with an expression for the partition function of the hypothetical fluid, permits evaluation of the properties of binary liquid mixtures by using only two adjustable binary parameters. Particular attention is given to the problem of nonrandomness in mixtures. A quantitative description of nonrandomness is achieved by combining the two-fluid concept with a hypothesis for ensemble averaging of a distribution of nearest-neighbor pairs.