Abstract: This paper studies the variance of blocked cross-validation estimator of the prediction error recently proposed in the literature. A more accuracy representation of the variance is provided and the main theorem shows that there exists no universal (valid under all distributions) unbiased estimator of the variance.