Abstract: Aiming at outlier mining in massive time series data, the paper combined grid clustering with MapReduce programming model to exclude grids that was impossible to contain outlier, and then used LOF Algorithm to detect outliers from the rest grids. In order to improve the detection accuracy of the grid clustering, this paper proposed an improved algorithm based on clustering radius. Experimental results showed the effectiveness of the improvement. Experiment also analyzed the execution time grid cluster cost under the circumstances with different number of nodes, which proved it was suitable for handling massive time series data combined MapReduce with grid clustering.