With the constant accumulation of data, there is much higher desire for processing and analysis power to handle these data. Since the traditional k-Nearest Neighbor (kNN) query algorithm is easy to cause load imbalance on account of the regular partitioning method and its current platform is single process or single machine platform which cannot obtain high enough overall performance today, an irregular partitioning method based kNN algorithm is presented and being executed on the distributed parallel computing model which positioning to process large scale datasets in a distributed parallel way— MapReduce in this paper. Experimental results and analysis show that the irregular partitioning method based kNN algorithm can realize much significant operational efficiencies and support efficient query of big data much better.