This paper is intended to discus the construction of granular logic system under variable fuzzy criterion condition and the formal representation of the reasoning of granular logic system so as to study the application of granular computing from another perspective and to facilitate the application of granular computing in artificial intelligence with uncertainty. The change law of fuzzy criterion variability can be described by the transformation of qualitative and quantitative attributes in attribute theory. On the basis that the concept and logic computing formula of attribute granule are given, a granular logic system is initially established and the logic reasoning is described using the formalization method of Petri Net, showing that the reasoning is effective and operations of thinking like cognition, recognition, judging and reasoning can be well expressed.