With the aim of avoiding faults during the construction of a software product, it is necessary to conduct tests from the early phases of this development. One of these phases is the software modeling. This paper presents a study related with faults detection during the learning process of modeling use case diagrams. To conduct this study, a controlled experiment was developed in which students who were completing their first software design course participated as subjects. The subjects used a proposed faults taxonomy. In this initial study, the effectiveness in fault detection in use case diagrams was evaluated. The result showed that by learning the common faults proposed in use case diagrams, it is possible to achieve moderately significant effectiveness (59.67% observed faults) in faults detection in this type of diagrams.