A novel nonlinear least squares frequency estimation method which uses only a one-dimensional search routine has been recently proposed in . In this paper, the variance of the estimator is analyzed at high signal-to-noise ratios. An analytical expression for the variance is derived. The dependence of the variance on the signal-to-noise ratio, the number of data partitions, the length of each partition and the frequency separation is investigated. Monte-Carlo simulation results are given to verify the analytical expression.