Metallurgical Digital Twin aims for minerals processing site-wide metallurgical and economical optimization of operations. It integrates mining data, consisting of run-of-mine ore characteristics, with a detailed mineral particle-based processing plant model. In addition to the feed ore properties, the Digital Twin model needs to be able to adapt into plant asset availabilities and operating constraints, as well as incorporating the operating costs and metal market price data for producing relevant KPIs for decision making. Adaptation of a processing plant model, based on the feed ore variations and equipment operating parameters and availabilities, implies need for connectivity with mining and real plant data sources. Vice versa, the results of a Metallurgical Digital Twin ‘what-if’ predictions are to be readily used for selecting the best control targets in closed loop manner. This approach for finding the global optimum is referred later on as the Plant Optimizer. Thus, these plant and site level targets generated by the Plant Optimizer are fed into optimizing controls of each process area, which are on top of stabilizing controls and process control systems to implement actual control actions. (Fig. 1).This paper describes concept and architecture of a Metallurgical Digital Twin that is based on a dynamic mineralogical predictive plant model. The Metallurgical Digital Twin enables predictions of short- and long-term process operating scenarios with alternative control actions and ore characteristics (Fig.2). As an example, a gold processing Digital Twin is presented. The process model is based on first principle equations and chemical reactions with empirical parameters. The flowsheet model was constructed with HSC Chemistry® software, with mineral processing and hydrometallurgy unit models (Fig. 3). In addition to technical aspects of a Metallurgical Digital Twin, some of the key drivers and success factors for a Digital Twin project are presented.