The Glass Box Test (GBT), also known as White Box Test or Structural Test, shows which parts of the program under test have, or have not, been executed. Many GBT tools are available for almost any programming language. Industry standards for safety-critical software require a very high or even complete coverage. At first glance, the GBT seems to be a well-established and mature testing technique that is based on standardized metrics. But on closer inspection, there are several serious shortcomings of the underlying models and metrics which lead to very imprecise, inconsistent coverage results of the various GBT tools. In this paper, a new and precise model for the GBT is presented. This model is used as a reference for the precise definition of all the popular coverage metrics that are around. The tool CodeCover which was developed in the University of Stuttgart is an implementation that strictly follows those definitions.