Icequakes at or near the bed of a glacier have the potential to allow us to investigate the interaction of ice with the underlying till or bedrock. Understanding this interaction is important for studying basal sliding of glaciers and ice streams, a critical process in ice dynamics models used to constrain future sea-level rise projections. However, seismic observations on glaciers can be dominated by seismic energy from surface crevassing. We present a method of automatically detecting basal icequakes and discriminating them from surface crevassing, comparing this method to a commonly used spectrum-based method of detecting icequakes. We use data from Skeidararjökull, an outlet glacier of the Vatnajökull Ice Cap, South-East Iceland, to demonstrate that our method outperforms the commonly used spectrum-based method. Our method detects a higher number of basal icequakes, has a lower rate of incorrectly identifying crevassing as basal icequakes and detects an additional, spatially independent basal icequake cluster. We also show independently that the icequakes do not originate from near the glacier surface. We conclude that the method described here is more effective than currently implemented methods for detecting and discriminating basal icequakes from surface crevassing.