Abstract : This thesis is submitted within the framework of formal description techniques used during the design process of real-time distributed systems. More precisely, it concerns the extension of the expression and analysis powers of the Stochastic Timed Petri Nets model (STPN). This model associates to each transition of a Petri Net a time interval and a density probability function on this interval (a lot of time characteristics can thus be represented and, in particular, time constraints). The expression power improvement concerns the introduction, on the one hand, of the age memory concept (which allows to study preemption mechanisms and is very useful in scheduling algorithms and dependability analysis) and, on the other hand, of several transition firing rules (in particular, MIN and MAX rules which allows o study worst cases, very important aspect in real-time systems). The different graphs which are obtained with this rules have been situated with respect to the behavioural reference which is the state class graph (considering only time intervals). The analysis power improvement concerns the definition of the Quantified Abstract Quotient Automaton concept (based on the Milner equivalence relation and Beizer rules) : it provides abstract models which are both qualitative and quantitative; il allows to control the size of the models to consider when modelling multilayer communication architectures. This model has been applied to the study of the communication protocol ARINC 629 CP (for plane systems). This study has demonstrated the real-time and fault tolerance properties of the protocol.