Reduced-order models are developed based on orthogonal collocation on finite elements for the steady-state optimization of multicomponent distillation units. The formulation divides the column sections into subdomains (finite elements) to track irregularities in the column profiles accurately and requires that plates with discontinuities in the flows introduced by feedstreams or sidestreams should be treated as discrete equilibrium stages. Application of the model formulation to a number of examples show that reduced-order models converge to the same optimal solution as the full-order, tray-by-tray model. Optimization results show that computational time can be reduced significantly using the more compact model formulation, and it is possible to converge to the optimal solution from a larger region for the initial set of variable values.