This work presents a mixed integer nonlinear programming (MINLP)-based superstructure optimisation approach to synthesize an optimal petroleum refinery topology or configuration for large-scale grassroots refinery systems. We develop a superstructure to include many possible prospective configurations and formulate rigorous models for the 32 commercial refinery processes that constitute the configurations, which gives rise to a convex MINLP model. The objective function is to maximize the total refinery profit for a given crude oil feed subject to material and energy balance constraints. We apply a two-level optimisation procedure: a master module to construct configurations from the superstructure and a submodule to optimize the process unit conversions and product temperatures of the configurations. A numerical example based on an actual operating refinery in Kuwait is illustrated to implement the approach with a resulting configuration that agrees with real-world practices. (C) 2019 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.