BACKGROUND Julesz named the ability of depth perception in random dot stereograms as "global stereopsis". At present there are only high-density random dot stereograms with different disparities in use in the ophthalmological practice. The aim of our exploration is to measure the effects on pattern recognition and depth perception by thinning out the dot density in random dot stereograms down to very low levels (< 1%). METHODS Experiments were performed by 43 volunteer observers using two sequences of random dot stereograms. Each sequence has its own constant disparity and within a sequence the dot density decreased to values less than 1%. Additionally we compared the performance in these tests with the performance in the conventional haploscopic stereoscopic tests (Titmus, TNO, Randot). RESULTS Of the observers 48.8% (n = 21) were able to establish a complete depth perception and pattern recognition ("global stereopsis") at a matrix dot density of 0.3% and a disparity of 730" (12, 17'). Another group of subjects (48.8%, n = 21) saw the sub-matrix at 0.3% dot density in depth as a star field ("local stereopsis") but was not able to recognise the form of the sub-matrix. A significant correlation does not exist between these results and those from the conventional stereoscopic test. CONCLUSIONS We assume from these results that the very low density random dot stereograms measure a new quality of stereoscopic vision, from which we can obtain information about the density of cortical binocular activated elements in the human brain.