This paper discusses foundations of conventional style of rule mining in which rules are extracted from a data table. Rule mining mainly uses the structure of a table, data partition, but two different approaches are observed: divide and conquer and covering: the former focuses on the nature of data partition and the latter does on the nature of information granules. This paper illustrates that granular computing gives a unified view of these two approaches, which may lead to theoretical foundations of data mining in the near future.