We use a nonlinear response formalism to describe the event plane correlations measured by the ATLAS collaboration. With one exception ($leftlangle cos(2Psi_2 - 6Psi_3 + 4 Psi_4) rightrangle$), the event plane correlations are qualitatively reproduced by considering the linear and quadratic response to the lowest cumulants. For the lowest harmonics such as $leftlangle cos(2Psi_2+3Psi_3 - 5Psi_5) rightrangle$, the correlations are quantitatively reproduced, even when the naive Glauber model prediction has the wrong sign relative to experiment. The quantitative agreement for the higher plane correlations (especially those involving $Psi_6$) is not as good. The centrality dependence of the correlations is naturally explained as an average of the linear and quadratic response.