In this paper extreme value theory (EVT) has been used to estimate the return levels for geomagnetic activity based on the aa index. The aa index is the longest, continuously recorded, geomagnetic dataset (from 1868 - Present). This long, 150 year, dataset is an ideal candidate for extreme value analysis. However the data are not independent and identically distributed as required for EVT since they are impacted by the approximately 11 year solar cycle. The Hilbert-Huang Transform has been used to identify the solar cycle component in the data and the data has been split into solar maximum and minimum times. In these two regimes the generalised extreme value distribution has been fit to the datasets. These have also been combined for an estimate of the overall return times. The results suggest that the largest event in the database (March 1989) is a one in 25 year event. However, considering separate solar maximum and minimum times has a large impact on the return times. During solar minimum conditions the return time of the March 1989 event is 130 years. This suggests that the occurrence of extreme space weather events is conditionally dependent on where in the Solar Cycle we are.