Abstract: In this paper, robustness of the best linear unbiased predictor in finite populations are investigated. Necessary and sufficient conditions for the predictor of linear predictable variable to be robust with respect to covariance matrices are obtained.