The major aim of this work is to introduce a new method for data reduction in information systems
classified by nonequivalence relations. The suggested approach is useful for many real life data
that can not be classsed into disjoint classes. The new approach is tested with examples and its
ability for decreasing the noisy in data, as well as simpliiing the structure for extracting decision
rules. Based on binary relation, this method is capable of discovering the reduct attributes or
indispensable attributes which are useful for data description andlor prediction, and to remove the
dispensable ones.