Indexing technique has been used extensively in order to facilitate and optimize query processing in various data retrieval and storage systems. Although an index technique can be used to reduce searching cost in both horizontal and vertical dimensions, it incurs both storage and maintenance costs. Index selection has been an active research subject, and various index selection methods have been given in literature. This paper will present a framework to mine frequent query patterns to select most frequently used access paths as candidates for index selection, and a Bayesian based method will be used to select index fields from the candidate set.