The introduction of appropriate crop rotations is known to be beneficial in many farming systems. One feature of rotations is that it takes a valuable length of time for the advantage of the rotation to take effect. In long-term rotation trials, the observations from the same plot over years are correlated; ignoring such correlations may affect the precision of the estimates of rotation effects. We examined five covariance structures between the plot errors over time to assess the effect of correlations on the standard errors of rotation means and rotation x cycle combination (interaction) means on wheat yields using eight years of data from six two-phase rotations with wheat. Based on wheat yield data from the four cycles of the rotations considered, the compound symmetry covariance structure (constant correlation) between plot errors arising over alternate years gave more efficient estimates of rotation means compared with the other four covariance structures.