Traditionally, optimization for large-scale multi-level lot sizing (MLLS) problems always encountered heavy computational burden. Scholars also indicated that "whatever the optimal method chosen to solve the MLLS problem, standard optimization packages were still faced with computer memory constraints and computational limits that prevented them from solving realistic size cases". Therefore, the main purpose of this paper is to propose an optimal method to reduce the computer memory while solving the large-scale MLLS problems. The optimal method is designed to implement on a database entirely because the demand for computer memory can be reduced significantly by means of the utilization of database storage. An example is given to illustrate the proposed method and computation capability is tested for the MLLS problems with up to 1000 levels and 12 periods. (C) 2010 Elsevier Ltd. All rights reserved.