Heat exchanger networks in process plants recover usable heat from hot to cold process streams such as feedstock, product, and pumparound streams. In this paper, we focus on the design of a heat exchanger network for removing variable heat from a complex fractionator in delayed coking units. A superstructure-based mixed integer nonlinear programming model is presented to minimize the total annual cost, including the heat exchanger area cost and the utility cost. In the proposed model, duties and operating parameters of pumparounds are optimized and the optimized structure of the heat exchanger network is defined simultaneously. Meanwhile, two modifications are applied to improve the simultaneous optimization model for the integrated system of the complex fractionator and heat exchanger network. First, a predominant calculation method for the log mean temperature difference is introduced in order to obtain the results closest to the actual process; second, nonconstant specific heat of process streams with pseudocomponents is considered to represent real thermal properties of process streams in the process model. A case study is carried out to demonstrate simultaneous optimization of the fractionator and heat exchanger network. The optimal design for the model is compared to the one with conventional design heuristics. By applying the proposed model to integrate the operation of the fractionator and heat exchanger network in a delayed coking unit, the total annual cost is reduced by 2.1 million CNY considering the constraints of the heat removals.