In this paper we present a new approach for the assessment of free-text answers. We use semantic networks, based on the MultiNet paradigm, as core for the computer-based analysis of short answers given by learners. In our assessment an answer is automatically transformed in MultiNet, then compared to a reference answer, and finally the results of this analysis are reformulated in natural language. Depending on this analysis and the user model a detailed feedback is presented to the learner for the improvement of his solution. We also describe the first use, advantages, limits, and perspective of our approach.