Argument evaluation, one of the central problems in argumentation theory, consists in studying what makes an argument a good one. This paper proposes a formal approach to argument evaluation from the perspective of justification logic. We adopt a multi-agent setting, accepting the intuitive idea that arguments are always evaluated by someone. Two general restrictions are imposed on our analysis: non-deductive arguments are left out and the goal of argument evaluation is fixed: supporting a given proposition. Methodologically, our approach uses several (...) existing tools borrowed from justification logic, awareness logic, doxastic logic and logics for belief dependence. We start by introducing a basic logic for argument evaluation, where a list of argumentative and doxastic notions can be expressed. Later on, we discuss how to capture the mentioned form of argument evaluation by defining a preference operator in the object language. The intuitive picture behind this definition is that, when assessing a couple of arguments, the agent puts them to a test consisting of several criteria. As a result of this process, a preference relation among the evaluated arguments is established by the agent. After showing that this operator suffers a special form of logical omniscience, called preferential omniscience, we discuss how to define an explicit version of it, more suitable to deal with non-ideal agents. The present work exploits the formal notion of awareness in order to model several informal phenomena: awareness of sentences, availability of arguments and communication between agents and external sources. We discuss several extensions of the basic logic and offer completeness and decidability results for all of them. (shrink)