Defining similarity metrics is one of the most important tasks when developing Case Based Reasoning (CBR) systems. The performance of the system heavily depends on the correct definition of its similarity metric. To reduce this sensitivity, similarity functions are parameterized with weights for features. Most approaches to learning feature weights assume CBR systems for classification tasks. In this paper we propose the use of similarity between case solutions as a heuristic to estimate similarity between case descriptions. This estimation is used to adjust weights for features. We present an experiment in the domain of Case Based Process Planning that shows the effectiveness of this approach.