Abstract: A partitioning algorithm is proposed to partition an entire hardware task into interconnected subtasks for reconfigurable computing.The algorithm，called PCGA，syncretizes probabilistic constructive （PC） algorithm and genetic algorithm （GA）.A new approach is proposed to measure the variety of partitions，and an initial population with a variety of better individuals is produced by PC algorithm.Then，the optimal solution is captured by GA based on this initial population.The experimental results show that PCGA can get better results of graph partitioning than those list-based partitioning algorithms，and use less runtime than those genetic algorithms based on a population of randomly generated individuals.