This paper reports the techniques of dynamic tuning of XML storage schema in VXMLR, which is a XML management system based on RDBMS. With two different tuning strategies, VXMLR can dynamically adjust its storage schema based on the latest query records to improve its query processing ef.ciency. When a tuning event is triggered, VXMLR first derives from its history queries the initial mapping rules that map XML DTD to relational schemas; then by vertically partitioning the relational tables or redundantly storing the data relevant to history queries, some candidate storage schemas are generated; following that, the benefit and cost of each candidate schema is estimated; and finally a cost-driven approach is proposed to select the final storage schema from the candidate schemas under a certain space constraint. Experimental results validate the practicability and effectiveness of the proposed techniques.