The Axis of Evil revisited

In light of the three-year data release from WMAP we re-examine the evidence for the ``Axis of Evil'' (AOE). We discover that previous statistics are not robust with respect to the data-sets available and different treatments of the galactic plane. We identify the cause of the instability and implement an alternative ``model selection'' approach. A comparison to Gaussian isotropic simulations find the features significant at the 94-98% level, depending on the particular AOE model. The Bayesian evidence finds lower significance, ranging from ``substantial'' at $\Delta(\ln E)\sim 1.4$, to no evidence for the most general AOE model.