Fundamental dynamic models of distillation columns are not well suited for incorporation into nonlinear model predictive control schemes due to their high state dimension. We investigate two reduced order nonlinear modeling techniques for high purity distillation columns. Compartmental and nonlinear wave models developed for a nitrogen purification column are compared to a rigorous dynamic simulator to assess the relative tradeoffs between prediction accuracy and computational complexity. The utility of each reduced order modeling technique for nonlinear model predictive control of high purity distillation columns is discussed.