This master thesis presents techniques for connecting requirements models expressed in a goal-oriented requirements engineering (GORE) paradigm into more operational specifications expressed in Event-B. More specifically, the objective was to produce a method that derives an Event-B model from a KAOS model, that relies on the semantics of those two languages, that guarantee a fine-grained traceability and that is as automatic as possible. After reviewing a number of existing approaches, none of those methods seem to answer the problem. Consequently an alternative approach was designed with the central focus of mapping GORE agents to Event-B machines. The work fully relies on the UML-B work for mapping data models. Recent Event-B extensions about machine decomposition are also used to decompose an initial system level machine into more finer grained agent machines based on their ability to control a specific piece of information. Finally the agent machines are refined to match the behaviour declared in the KAOS model. The approach has been partially implemented in a prototype that uses model to model transformation technologies (EMF - ATL), and has been validated on different cases.