The process of enriching the C-13 isotope, performed in trains of cryogenic distillation columns, exhibits large settling times, nonlinearities, large dead-times, and are difficult to model precisely. Such equipment has been developed in Romania, with concentration increasing up to 70 %. A control analysis for a single unit has already been done including a decentralized multivariable PI controller and two decoupling control algorithms based on the internal model control (IMC) approach. Here, a multivariable predictive controller, the extended prediction self-adaptive controller is proposed. The simulation results, considering significant modeling errors, demonstrate that this represents a more suitable choice than the previously designed strategies. Comparisons are included to support this idea.