International audienceIn our previouswork (Dillon and Mansour 2009), a stochastic reliability model of atomic web services was proposed. Using the well-known classic two-state bounded set technique, we developed a service-oriented model that dynamically calculates the reliability of composite web services with rollback recovery (Mansour and Dillon in IEEE Trans ServComput 4(4), 2011). In order to improve theQuality of Service, fault tolerance techniques have been introduced using recovery block adaptation. Our workflow was based on series-parallel structures that constitute parts of existing structures. It is worth mentioning that major serviceoriented systems contain larger and more complex structures than the simple series and parallel ones. This is a limitation in our previous approach. In order to consider more realistic service-oriented systems, other main structures, such as AND, XOR and Loop, should be included into our model. In this article, our previous structures are generalized to include AND, XOR and Loops. In addition to generalized structures, we extended the existing two-state bounded set technique to include three-state systems. This extension was especially motivated by XOR-based structures. A comparative study between bounded set techniques and a new stochastic model is also presented. Our simulation results accurately reflect the performance of the new proposed model and confirm our theoretical studies. Furthermore, Monte Carlos simulations were performed and the results obtained clearly validate our stochastic model