In some of the papers on Data Envelopment Analysis (DEA), it has been explained that if correlation coefficient between each pair of input (output) vectors is strong and positive, one of the input (output) vectors should be omitted. In this paper, a threshold is identified for correlation coefficient beyond which one of the input vectors could be omitted without altering the mean efficiency of the DMUs. The threshold identification in terms of some of the DEA models including CCR, CCRCSW, BCC and BCCCSW is performed. First, the effect of the factor of before and after omission of one of the input vectors on the mean efficiency is determined by analysis of variance (ANOVA). Then, the effect of the factor of different correlation coefficients between two input vectors on the mean efficiency is determined. Third, interaction effect of the factor of different correlation coefficients between two input vectors and factor of before and after omission of one of the input vectors on the mean efficiency is determined. Finally, the interaction effect of the factor of different correlation coefficients between two input vectors and factor of size of DMUs on the mean efficiency is determined.