Scarcity of freshwater resources and increasingly stringent environmental regulations on industrial effluents have motivated the process industry to identify and develop various water recovery strategies. This work proposes the use of detailed model representation for water regeneration network synthesis, in which nonlinear mechanistic models of the regeneration units are embedded within an overall mixed-integer nonlinear programming (MINLP) optimization framework. The superstructure-based MINLP framework involves both continuous variables for water flow rates and contaminant concentrations and 0-1 variables for selection, of piping interconnections. The nonlinear regeneration model produces a rigorous cost-based relation, instead of a "black box" model, that is incorporated within the overall MINLP representing a network of numerous water sources and water sinks. Hence, such an approach enables a simultaneous evaluation of both direct water reuse/recycle and regeneration reuse/recycle opportunities. To demonstrate the proposed approach, an industrial case study is illustrated that incorporates a mechanistic model of reverse osmosis network (RON) for water regeneration for an operating refinery in Malaysia. The results indicate a potential of 58% savings in freshwater use. The capital investment for the water regeneration network is reported as $8,960,000 with a payback period of 2.1 years, thus providing economic support to pursue the RON retrofit option.