Abstract: Data reduction is one of the important topics in knowledge discovery in databases. Attribute oriented generalization （AOG） can be used in data reduction. First, AOG and its shortcomings are discussed from the view of data reduction: the single attribute threshold control of AOG is boolean control, so when there are exceptions, it can cause data over reduction and lose the meanings of data reduction. Then, quantifiable attribute oriented generalization （QAOG） based on AOG is presented in order to develop AOG: record threshold is introduced; attribute threshold and record threshold are used simultaneously in order to change boolean control to quantifiable control, so when there is no exception, it has the same results as AOG, otherwise, it has better results than AOG. An efficient algorithm of QAOG is also given.