Recently,trust emerged as a momentous aspect to evaluate resources, services or persons. In our work, the trust notion focuses on a system as a whole and from the point of view of a particular user to do a particular digital activity as editing a document, mailing, chatting, etc. Our general goals are (i) to enable users to have a personal comparison of applications allowing them to do an activity such that they can choose the one satisfying their personal expectations and (ii) to know how trustworthy their system is to do a particular activity (all applications together). We consider a system as a graph composed of paths where the source is a person and the target is a final application or data. We consider that trust in a system depends on its architecture and we identify two problems (i) how to evaluate trust in a graph having dependent paths i.e., paths having com- mon nodes, and (ii) how to express and deal with uncertainty in evaluating trust in a system. Concerning the first problem, trust approaches based on graphs have been proposed in the domain of social networks. Their solution for dependent paths is either removing paths or just choosing one of them what causes loss of information. Considering the second problem, subjective logic emerged to express trust as a subjective opinion with a degree of uncertainty. In this paper we present SUBJECTIVETRUST, an approach that relies on subjective logic to evaluate trust in distributed systems. It proposes two solutions to treat dependent paths and takes into account the shape of the system architecture in trust evaluation. We analyze SUBJECTIVETRUST in a series of experiments that show its accuracy.