In order to build Grid resource sharing environment, a basic problem is to locate one or more entities (e.g. CPU, memory, OS, applications) that match a provided description, this called resource discovery process. This paper proposes a model of resource discovery method that based on classifying the nodes of resources according to their importance to a Grid application. Each group has a leader that collects the information of the group members. The paper describes how this model can satisfy the expected characteristics of any resource discovery method. Which are: decentralization, scalability, dynamism, supporting attribute-based discovery and multi-attribute range queries. © 2008 IEEE.