Traditional rule learners employ equality relations between attributes and values to express decision rules. However, inequality relationships, as supplementary relations to equation, can make up a new function for complex knowledge acquisition. We firstly discuss an extended compensatory model of decision table, and examine how it can simultaneously express both equality and inequality relationships of attributes and values. In order to cope with large-scale compensatory decision table, we propose a scalable inequality rule leaner, which initially compresses the input spaces of attribute value pairs. Example and experimental results show that the proposed learner can generate compact rule sets that maintain higher classification accuracies than equality rule learners.