As far as we know, a little research of mining global maximum frequent itemsets has been done. The paper proposed an algorithm for mining global maximum frequent itemsets based on FP-tree, namely, AMGMFI algorithm. AMGMFI algorithm makes computer nodes computed local maximum frequent itemsets independently with DMFIA algorithm, then other computer nodes exchanged data with the center node, finally, global maximum frequent itemsets were gained. AMGMFI required far less communication traffic by the search strategy of top-down. Theoretical analysis and experimental results suggest that AMGMFI algorithm is effective.