Distillation column has multivariable and nonlinear characteristics. High operation cost of distillation column required energy consumption optimization. The new alternative method to find out thelowestenergy consumtion of distillation column is optimization method using genetic algorithm. In this research, distillation model built up by neural network Multi Layer Perceptron (MLP) with Nonlinear Auto Regressive with eXternal input (NARX) structure, learning algorithm using Levenberg-Marquardt. Neural Network model has RMSE 3.9974x10-4 for condenser duty and RMSE 1.7435x10-4 for reboiler duty. Genetic algorithm optimization results are Qc 1.85E+07 and Qr 1.05E+07 which process variables are top pressure 106.846 Kpa, level condenser 30.289%, temperature feed 83.48 oC, fraction feed 0.5258, flow feed 493.518Kgmol/hour. In other word, there are decreasing steam and cooling water cost up to 46.2 %.