There are two major challenges to the personalized recommendation method, one is the sparseness of characteristic attribute, the other is the excessive reliance on scoring data. To solve above problems, a personalized recommendation algorithm (PRM-Grey) based on grey theory is presented. Firstly, the nearest neighbor matrix formed through the similarity between the characteristic matrix rows. Then, PRM-Grey combines with the corresponding data scores make recommendations. It can effectively solve characteristic attribute spares of personalized recommendation. On this basis, PRM-Grey imports grey relational analysis to measure the similarity of characteristic matrix, and uses grey prediction model to make personalized recommendation. Experiments show: compared to traditional personalized recommendation method, the accuracy of the PRM-Grey gains an average 10%. It fully illustrates the effectiveness of PRM-Grey.