Abstract: Imputing missing values is one of the challenges in data mining and machine learning.Missing values in a dataset can decrease the efficiency of learning algorithm and negatively affect the algorithm.Existing imputation methods for missing values can not fully satisfy the users’ increasing requirements.In this paper，a novel nonparametric algorithm is proposed by using the gray system theory.In this algorithm，missing values are imputed iteratively until the algorithm converges or the output matches to the users’ requirement.Experiments with the UCI dataset demonstrate that our method performs better than many existing algorithms such as the KNN algorithm and the mean method in terms of imputation efficiency.