A new perimetric pattern discrimination test was compared with conventional automated perimetry (Humphrey program 30-2 or Octopus program 32) in glaucoma patients, glaucoma suspects, and control subjects. The new test is based on the rationale that a greater percentage of retinal ganglion cells should be needed to detect a stimulus by its shape, or pattern, than by its brightness. The pattern discrimination stimulus was apatch of nonrandom dots embedded in a surrounding random dot field of the same average density. Pattern discrimination thresholds were measured by changing the degree of regularity, or coherence, of the stimulus dots. The fully coherent target was a static, 1-s duration, 20 x 20-dot checkerboard. Using a criterion-free relative operating characteristic analysis, we estimated the ability of both the pattern discrimination and conventional tests to distinguish the normal data distribution from the suspect and glaucoma distributions. The pattern discrimination test appeared to produce separations greater than conventional perimetry for glaucoma suspects and separations equivalent to conventional perimetry for glaucoma patients.