Methods Observers saw L1 followed by L2 (ISI = .3 ... 1 s); L1 was a rectangular lattice, with variable aspect ratio r1, and aligned with a randomly rotated system of polar coordinates so that its two most likely perceptual organizations were parallel to 0 ° and 90 °. L2 was a square lattice (r2 = 1.0) with two most likely perceptual organizations parallel to 0° and 90°, and was rotated relative to L1 by 0° <= q <= 45° (Dq  = 15°). Observers reported on the organization of L1 and L2. [Expreression Ln -> y means that Ln was grouped parallel to y°, and p(Ln -> y) stands for the probability of this event. ]

Results p(L1 -> 0) increased with r1, but p(L2 -> 0) was unaffected by r1 (regardless of q).  However, conditional probabilities p(L2 -> 0 | L1 -> 0) and p(L2 -> 0 | L1 -> 90) decreased with r1 for low values of q, becoming negligible for q >=  30° (see the Figure).

Conclusions Internal noise, which is independent of the stimulus, is additive with the effect of r1.  The noise explains why the organization of L1 is not always predictable by r1.  The persistence (>= 1 s) of the noise is evident with low r1 for p(L2 -> 0 | L1 -> 0), and with high r1 for p(L2 -> 0 | L1 -> 90). The noise is orientation-tuned; the smaller q the higher the correlation of the noise samples that are added to the signals of L1 and L2.