Efforts to derive value-added chemicals from lignin are hindered in part because of its great structural complexity and heterogeneity. Lignin is a polydisperse, random copolymer with varying proportions of monomers and interunit linkage types, forming a hyperbranched topology. Structural models comprising a single macromolecular representation have been proposed for specific biomass sources; these representations, although consistent with experimental observations, are not unique. Alternative models consisting of populations of representations whose average properties match experiment have also been proposed. Expanding on this latter approach, we have developed a stochastic method of generating libraries of diverse and complex structural representations of lignin that is generally applicable to any biomass source. For demonstration purposes, the method is applied to create libraries of wheat straw lignin, a representative herbaceous lignin. The libraries are statistically validated through the chi(2) goodness-of-fit criterion and Student's t test to conform to the following properties that were measured experimentally: monomer, bond, and molecular weight distributions and branching coefficient. Novel model predictions include the distribution of aromatic hydroxyl groups, moieties that are known to accelerate the thermochemical deconstruction of lignin, and the dyadic bonding distribution, which could aid efforts to understand the causes behind the observed biological activity of lignin deconstruction products. The robustness of the method is demonstrated by exploring the run-to-run variability, thus strengthening predictions made by subsequent utilization of the libraries.