An　improved　circuit　partitioning　algorithm　based　on　the　greedy　randomized　adaptive　search　procedure　(GRASP)　was　presented　to　improve　the　circuit　partitioning　quality　of　traditional　iterative　improvement-based　algorithms.　GRASP　consisted　of　a　construction　phase　and　a　local　search　phase　and　could　construct　good　initial　partitions　quickly.　In　the　construction　phase,　a　heuristic　strategy　was　introduced　to　select　clusters.　A　very　efficient　searching　technique　called　Path-Relinking　was　integrated　into　the　GRASP　iterative　process　to　build　paths　among　local　optimal　solutions　and　effectively　explore　the　local　optimal　solution　space.　The　experimental　results　indicated　that　compared　to　the　basic　GRASP,　the　modified　algorithm　improved　the　solution　quality　in　a　reasonable　period,　and　obtained　better　partition.　The　minimum　cut-size　reached　33.3%　while　the　average　cut-size　was　up　to　27.4%.