Metacognitive strategies are needed to solve complex problems such as mathematical modelling problems in groups in a goal-oriented and successful manner. In this paper, the results of an intervention study aiming at enhancing students metacognitive group strategies for modelling are presented. Students from 18 different classes (grade nine and ten) worked on six different modelling problems over a period of 10 months. In a pre-post-test design, their metacognitive group strategies were measured by self-reports. Data analysis shows that up to 33% of the total variance of items lay between groups. Thus, a multilevel analysis was conducted with the student level as within level and a group level as between level. The comparison of different models revealed that the assumed structure of metacognitive strategies, divided into the three factors of strategies for proceeding, for regulating and for evaluating for the student level, can be assumed for the group level as well. In addition, the results show that the teaching unit resulted in a significant increase in evaluation strategies in the intervention group, but not in the control group, and not according the other kinds of strategies. Hypotheses to explain these outcomes are established in this report.