Mathematical modeling describes the processes being performed by people who try to solve real world problems mathematically or who try to solve mathematical word problems. The literature on mathematical modeling is based on the so-called modeling cycle (Blum, 2003), which consists of five substages, that describe the course of the problem-solving process. As a first step (Structuring), the problem situation is being understood by the modeler and the most crucial elements of the problem are identified. The modeler then translates those elements to a mathematical language (Mathematising), where they are now presented in the form of equations, numbers, operators and symbols, i.e. in the form of a mathematical model. In the third step (Solving), the model is used to come to a mathematical solution to the problem, which in the next step (Interpretation) is then re-translated into real-world information and gives an answer to the question that has been raised in the problem. The last step describes the Validation of the whole problem-solving process. If the modeler comes to the conclusion that his solution is wrong or not optimal, the modeling cycle starts from the beginning. The modeling cycle functions as a descriptive model of modeling processes, because it describes how modeling is carried out by students. On the other hand, it also functions as a normative model, because it makes prescriptions on how one should behave, when solving problems mathematically. However, there do not exist any empirical studies to prove neither the descriptive nor the normative character of the modeling cycle. This thesis shows, that the suitability as a descriptive model can only be affirmed with restrictions. Validation could not be observed in any of the subjects taking part in the first study and the sequence of the remaining four processes also diverged from the theoretically postulated order. A revised version of the modeling cycle as a recursive model is therefore proposed. In the second study, the suitability as a normative model was explored, by separately assessing the four processes Structuring, Mathematising, Solving and Interpretation with newly designed items. The four scales were validated with several criteria, one of them the ability to solve modeling tasks, which confirms the normative function of the modeling cycle. In order to help promote modeling competency in young learners, it is necessary to find ways for diagnosing modeling competency as early as at the end of primary school. By generating items for the four sub-competencies Structuring, Mathematising, Solving and Interpretation a basis has been founded to do this in an objective, reliabel and valid way.