Classic query optimization techniques use metadata (e. g. statistics) to estimate the cost of di erent possible solutions (query plans) to evaluate a query and choose the one that minimizes a cost function. This cost function determines the optimization objective, which is generally the evaluation time. In ubiquitous computing environments, where devices are autonomous and have limited physical characteristics (e.g. energy or CPU power), the evaluation time of queries is no longer the main optimization objective. Moreover metadata required for a priori estimating the evaluation cost of query plans are not always available. In order to solve this problem, this article describes a new query optimization technique based on automatic learning, particularly on case-based reasoning. This technique is not another cost function-based query optimization technique. It considers a query case base where cases represent experiences of queries that have been previously executed, optimized and evaluated. It address the exploitation of the case base for generating execution plans and learning on existing ones according to measures inserted in the case base. Currently we are working in the prototype that implements the query optimization techique that we propose.