We study estimation correlation from bivariate data separated vertically, i.e., component-wise into two servers with privacy constraints.