A novel mathematical programming method was applied to acquire the optimal heat exchanger network (HEN) retrofit in the crude oil distillation unit. The mathematical model was based on the stagewise superstructure model with splits of streams from the work of Yee et al. [Chin. J. Chem. Eng. 2007, 15 (2), 296-301]. The objective function was developed into the minimum of the sum of the cost of modifications and the cost of utilities in one year, focusing on the changing heat transfer coefficients arising in the reused existing heat exchangers from stream splits. The HEN modifications included buying hew heat exchangers, adding or reducing areas of the existing heat exchangers, moving the existing heat exchangers, and laying pipes. It was a mixed integer nonlinear program (MINLP) problem. A hybrid genetic algorithm was used to optimize the MINLP problem. In order to reduce the total cost of modifications, the heat exchangers and the corresponding matching structures between the hot and cold streams in the existing HENs were designed as elites introduced into the hybrid genetic algorithm. The Dittus-Boelter and Kern correlations were used to estimate the convective heat transfer coefficients of the corresponding streams. One example was given to show the effect of the novel retrofit strategy.