Tray efficiencies are data fundamental to the evaluation of the performance of distillation columns, and many estimation methods have been proposed on the basis of the mass-transfer process. However, determining tray efficiencies in industrial practice remains a challenge. In the current paper, we provide a new method of estimating tray efficiencies of distillation columns by using industrial operating data, in which a multiobjective optimization problem is formulated and a univariate marginal distribution algorithm is used to solve this problem. Using the obtained tray efficiencies, a neural network model is then trained to predict the future efficiencies. The feasibility of the proposed method is verified by an industrial C8 aromatics distillation column. With the prediction errors limited to 5%, the approach shows more accurate generalization than traditional methods.