1. For 12 subjects (4 male, mean age 24.1 years) the EEG was recorded from 19 Ag-AgCl electrodes (electrode positions are shown in Figure 1[a]) according to the international 10–20 system. Subjects performed a variant of the primed target detection task described in Elliott and Müller (1998). The experiment described here employed a priming-display presentation frequency of 40 Hz while priming displays were presented for 600 msec and followed immediately by target-display presentation. The electrodes were mounted in an elastic cap, were referenced to Fz while the nose served as the ground electrode. Electrode impedance was maintained below 5 kOhm. Horizontal and vertical electrooculargrams (EOG) were additionally registered with four electrodes. EEG activity was amplified by means of NeuroScan amplifiers, digitized on-line with a sampling rate of 500 Hz and analog-filtered with a 0.1-Hz high-pass and a 100-Hz low-pass filter. A 50-Hz notch filter was applied to remove artifacts related to the main's electricity supply.

For the recording of EOG, the time constant 300 msec with a low pass filter at 70 Hz was used. The EOG-channel was visually inspected for each trial, and trials with eye movement or blink artifact were rejected. Localized muscle artefacts (at electrodes T3 and T4) were identified and if present reconstructed by means of an extended independent components analysis (ICA) algorithm (see Makeig et al. 1999). Averaging epochs lasted from termination of an alerting tone 200 msec before until 1,200 msec after priming-display presentation. Baselines were computed in the – 200 to 0 msec interval for each trial and subtracted prior to subsequent analyses. Analyses were carried out on the averaged event-related potential (ERP) for each subject.

In a first step, a series of component activations were recovered from each averaged signal by means of ICA using information maximization (infomax) techniques described by Bell and Sejnowski (1995) with variants of the ICA Matlab package (v.3.52) (available at: http://www.cnl.salk.edu/~scott/). In order to classify components and identify particular groups of clusters that appeared during premask-matrix presentation, components were defined in terms of the latency and topographical distribution of variance maxima (in this case, topographical projections were standardized by substituting raw activation at each electrode with the corresponding zvalue computed relative to all projected activations at the time of maximal activation). Classification then proceeded by means of cluster analysis, calculating Euclidean distance between objects and computing linkages in a hierarchical cluster tree based upon the average distances between groups of objects and a threshold of 19 clusters (cophonetic correlation coefficient c = 0.81). The resulting clusters were considered for further analysis if (i) they included activations from more than 75% of subjects (i.e., 9 or more of 12 activations), (ii) they were specific to priming-stimulus presentation, (iii) maxima fell within the period of priming-display presentation, and (iv) if, following examination of the frequency component of each component activation by means of a 256-point fast-Fourier transform (FFT), strong peaks were evident at, or close to the priming-display presentation frequency of 40 Hz. On these criteria, a single component cluster was identified, which is described in Figure 1 and the main body of text.