Symmetry is a highly salient feature of animals, plants, and the constructed environment. Although the perceptual phenomenology of symmetry processing is well understood, little is known about the underlying neural mechanisms. Here we use visual evoked potentials to measure the time course of neural events associated with the extraction of symmetry in random dot fields. We presented sparse random dot patterns that were symmetric about both the vertical and horizontal axes. Symmetric patterns were alternated with random patterns of the same density every 500 msec, using new exemplars of symmetric and random patterns on each image update. Random/random exchanges were used as a control. The response to updates of random patterns was multiphasic, consisting of P65, N90, P110, N140 and P220 peaks. The response to symmetric/random sequences was indistinguishable from that for random/random sequences up to about 220 msec, after which the response to symmetric patterns became relatively more negative. Symmetry in random dot patterns thus appears to be extracted after an initial response phase that is indifferent to configuration. These results are consistent with the hypothesis (Lee, Mumford, Romero, & Lamme, 1998; Tyler & Baseler, 1998) that the symmetry property is extracted by processing in extrastriate cortex.

^{2}) presented on a dark background (5 cd/m

^{2}) at 15% dot density. These fields were calculated online and drawn to video memory during vertical refresh. Symmetry was introduced into the patterns by reflecting the upper left quadrant of the pattern about the vertical axis and then reflecting the top pattern about the horizontal axis, thus creating two-fold symmetry about the horizontal and vertical axes (see Figure 1 for an illustration). New patterns, either symmetric or random, were presented every 500 msec, with the pattern remaining continuously visible for 500 msec between image updates. By redrawing the patterns on every transition, we avoided the possibility that the response is determined by a particular feature of any individual pattern. In a typical experiment, the observer was presented with 60 to 100 different exemplars of symmetric patterns and a comparable number of random patterns.

_{1}, O

_{z}, and O

_{2}, each referenced to C

_{z}. The skin was prepared with Omni-Prep, and 10–20 conductive cream (D.O. Weaver) was applied. Electrode impedances were between 3 and 10 kilo-ohms. The electroencephalogram was amplified 50,000 times with Grass Model 12 amplifiers and digitized to 16 bits accuracy at a sampling rate of 434 Hz. Analog filter settings were 0.3 to 100 Hz, measured at −6 dB points.

_{z}versus C

_{z}) are shown in Figure 2. The main response consisted of a biphasic potential with an initial positive peak at 100 msec followed by a negative peak at 145 msec. There are two responses per epoch, and these responses are nearly identical (Red trace). Time-averaged responses to a symmetric-random sequence for the same observer are shown in Figure 2, Blue trace. The initial transition is from random to symmetric (symmetry onset), and the second transition at 500 msec is from symmetric to random patterns (symmetry offset). The initial activation with symmetric/random sequences is the same shape as that for random/random updates (compare Red and Blue traces in Figure 2). However, the response to symmetric/random sequences becomes more negative starting at 200 msec after a random-to-symmetric transition and more positive starting at 165 msec after transitions from symmetric to random patterns.

*p*< 0.0001; paired

*t*test). In contrast, the amplitudes at the second harmonic did not differ (1.68 ± 0.65 and 1.39 ± 0.64 microvolts;

*p*= 0.76) for the symmetry random versus random/random cases. Amplitudes at the 8th harmonic, which are representative of the higher even harmonics, also did not differ in the two conditions (1.19 ± 0.41 and 1.11 ± 0.39 microvolts;

*p*= 0.89).

^{th}Annual Vision Research Conference, April 28–29, 2000, Ft. Lauderdale, FL.