Another technique is the correction by using a dipole model [4]. With this multiple source eye correction (MSEC) method ocular artefacts are modelled by moving dipoles of the eyes, and this activity is subtracted from the EEG.
http://plosone.org/article/info:doi/10.1371/journal.pone.0003004#article1.body1.sec1.p3

At this point one citation is missing:

Berg P, Scherg M (1994). A multiple source approach to the correction of eye artifacts. Electroencephalogr Clin Neurophysiol 90: 229-241.

Further, the method is not described correctly as stated by P. Berg in the comments section:

"Eye activity is estimated from some estimate of the topography of eye movements, for example, by averaging eyeblinks and then using the first component of a spatial PCA. Note that regression from eye channels to the EEG could also be used to estimate the topographies!
Brain activity is modeled either by components of a spatial PCA applied to sections of the EEG not containing eye activity ("adaptive" method), or by a dipole model ("surrogate" method). The result is a matrix of topographies combining eye and brain topographies.

Using the methods described by Scherg and others for spatio-temporal source analysis, the eye activity waveforms are then estimated by multiplying the pseudoinverse of the matrix onto the data."