This study outlines an adaptive method which constructs improved query vectors based on the user preference judgments on sample document pairs. In particular, the user states that some documents are preferred to other documents and the system is then expected to rank the preferred documents ahead of the others. In the adaptive system, all needed parameter values are provided within the model, and a solution query vector is constructed under well defined conditions. Certain relationships between the new adaptive and the conventional relevance feedback systems are discussed and evaluation data are provided to demonstrate the effectiveness of the system.