We determined the spatial parameters that described the visual detection of spatio-temporal correlation in moving two-dimensional noise patterns. The target field (5.21 x 5.31 deg of visual angle) was divided into horizontal stripes of equal width D. Adjacent bars alternately contained noise patterns moving with velocity leads to V1 and leads to V2. We varied D, leads to V1 and leads to V2. Roughly three different percepts occurred. If the stripes were very broad the different movements in alternate stripes were perceived together with the division of the field into stripes. If the stripes were very narrow the division into stripes was not seen, but the moving noise patterns with velocities leads to V1 and leads to V2 were perceived as transparent sheets moving through each other. For intermediate stripe widths the target field looked incoherent and the subject was not clear about the percept. In this region the subject found it difficult and sometimes impossible to discriminate these patterns from a completely uncorrelated spatio-temporal white noise pattern (snow). To quantify the detectability, the patterns were masked with snow (spatio-temporal white noise). The r.m.s. contrast of the total stimulus was kept at a constant value, whereas the subject set the signal-to-noise ratio (SNR) to a threshold value. At certain bar width the thresholds reached a maximum value. Thse critical bar widths depended on the velocities leads to V1 and leads to V2. These critical bar widths were interpreted in terms of a simple general model for the detection of spatio-temporal correlation. In these terms the span of the elementary correlators rose monotonically with the velocity to which the correlator is most sensitive.