The cross‐correlation function (CCF) and autocorrelation function (ACF) for Great Lakes annual net basin supplies are derived under the assumptions that annual lake outflows and water levels are autoregressive lag 1 processes and that each lake behaves as a linear reservoir. Except for the pairing between Lakes Superior and Michigan‐Huron, there is reasonable agreement between the sample and predicted CCFs, especially among net basin supplies to the small lower basin lakes. The derived ACFs reduce to an expression identical in form to the ACF for an ARMA(1, 1) process at all the Great Lakes except Lake Michigan‐Huron. At the upper basin lakes, sample and predicted ACFs drop to zero rapidly. At the lower basin lakes the ACFs exhibit a much more gradual decay, suggesting the presence of long‐term persistence. Prominent tails in the ACFs of the annual net basin supplies have been attributed to historical shifts in the precipitation regime at the lower basin lakes. Results from this study show that the residual method currently used to estimate net basin supplies can also induce a similar artificial long tail in the ACF. This observation has important ramifications in efforts to simulate Great Lakes water levels, since simulation results are quite sensitive to the covariance structure of the annual net basin supplies.