Rough set theory is based on information granules. This paper studies information granules based on a decision logic language in information tables. In this paper, the theorems of determining definable granules and definable partitions are given. Furthermore, this paper gives the definitions of the definable upper and lower approximations of indefinable granules, and studies their properties. Through the descriptions of the definable upper and lower approximations, we propose a way of describing indefinable granules. As a result, we can obtain some explicit and useful information on indefinable granules. This is then an approach to discover knowledge hidden in indefinable granules.