Valid identification of blink artefacts: are they larger than 50 μV in EEG records?

Can artefacts, in particular of blink potentials, be validly identified in EEG data by defining EEG amplitudes as artefacts whenever their absolute values exceed 50 μV? Does the performance of this 50 μV criterion change when the data have been high-pass filtered (simulating a low time constant)? These questions were studied in data of an auditory oddball task recorded from young and elderly healthy adults and from Alzheimer patients. The performance of the 50 μV criterion heavily dependend on the distance from the eyes: most blinks were detected at Fz, very few at Pz and Oz. This rate further decreased after high-pass filtering. A qualitative effect of the 50 μV criterion occurred in the Alzheimer patients' Pz data: unidentified blink artefacts caused a late positive wave that mimicked a delayed P3. In the high-pass filtered data, this effect occurred not only at Pz but also at Cz. These results lead to the conclusion that the 50 μV criterion does not in general perform well, and in particular may bias results when records are made from Pz or from Cz only using a short time constant.