Conformance testing is the process of checking that a system possesses a set of desired properties and behaves in accordance with some predefined requirements. In this context, passive testing techniques are used when the system under test cannot be interrupted or access to the system's interfaces is unavailable. Passive testing relies on the observation of the implementation during runtime, and the comparison of the observation with the expected behavior, defined through conformance properties. The objective of this thesis is to define a novel methodology to validate communicating protocols by passive testing. Existing approaches are derived from works with finite-state and labelled transition specifications and as such, they presume there exists a causality relation between the events observed in the implementation (the trace). When dealing with message-based protocols, such as the Session Initiation Protocol (fundamental for IMS services), such causality does not necessarily exist and furthermore, it may only be determined through data parts. Since existing techniques are optimized for dealing with control parts, they present limitations for testing based on data parts: reduced expressibility and succinctness of conformance properties, as well as problems to deal with satisfaction of properties including future conditions. In this work we present a message-based/data-centric approach for dealing with these issues. Observations in a trace are in the form of messages. Expected behavior is defined in a bottom-up fashion, starting from expected criteria that must be fulfilled by one or more messages, defined as constraints between the message data fields. Temporal relations by quantification over the criteria, e.g. a property may require that certain criteria "must be held for all messages in the trace". Our approach allows to express formulas about the future and past of the trace, allowing to define more general criteria than through control parts alone. Issues related to satisfaction of properties and declaration of conformance verdicts are also discussed here. Although observation of a behavior defined as a property is indication of conformance, lack of observation is not necessarily indicative of a fault. Several solutions to this issue have been proposed and implemented in this work. Finally, our work presents interesting perspectives, in terms of extensibility for online detection or improved expressiveness, but also since a message-based approach provides an alternative view to traditional testing techniques
