A systematic framework is presented for the representation of superstructures and derivation of optimization models in process synthesis. The state task network (STN) and state equipment network (SEN) are proposed as the two fundamental representations of superstructures for process systems involving mass, heat and momentum transfer. The mathematical modeling of either of the two representations is performed with generalized disjunctive programming (GDP), and then converted systematically into mixed integer linear programs/mixed integer non-linear programs (MILP/MINLP) problems. The application of this methodology is illustrated with the synthesis of distillation sequences, with and without heat integration, which lead to MILP problems. It is shown that ad hoc models that have been reported in the literature can be systematically derived, and in the case of separation sequences with heat integration, a new improved model is derived. Numerical results for comparing alternative models are also presented.