Abstract: Since Grey Relational Analysis (GRA) can measure the similarity between the reference samples and the compared samples, it is widely applied in clustering and classification, especially, on the condition that the information of samples is incomplete and the size of samples is small.In general, the time domain of gray relation is the data of the horizontal features of each sample.Differed from the traditional GRA, a novel aspect is revealed, where the time domain of gray relation becomes the data of the vertical features of each sample.After defining the reference vector by ranking or labels in classification, grey relational analysis is conducted between the corresponding column of each feature and the referenced vector, thus, the feature importance is computed by the grey relational grade.Then feature selection can be performed with the feature importance, so as to conduct dimension reduction for sparse and large number of features.Worthy to be pointed is that the presented method can provide the parameters of features importance with good interpretation ability.To verify the importance evaluation method, experiments are performed on Decathlon and IRIS data set and the experimental results show that it is consistent to the priror knowledge.