Query optimization is a widely studied problem, a variety of query optimization techniques have been suggested. These approaches are presented in the framework of classical query evaluation procedures that rely upon cost models heavily dependent of metadata (e.g. statistics and cardinality estimates) and that typically are restricted to execution time estimation. There exist computational environments where metadata acquisition and support is very expensive, additionally; execution time is not the only optimization objective of interest. An ubiquitous computing environment is an appropriate example where classical query optimization techniques are not useful any more. In order to solve this problem, this article presents a query optimization technique based on learning, particularly on case-based reasoning. Given a query, the knowledge acquired from previous experiences is exploited in order to propose reasonable solutions. It is possible to learn from each new experience in order to suggest better solutions to solve future queries.