The extraction of a global orientation structure presumably has a different neural mechanism from that of the analysis of its local features. We investigated spatial integration within these two mechanisms using stimulus patterns composed of dot pairs (dipoles). The stimuli targeted local feature detection, contained no global configuration, but rather contained randomly oriented dipoles of a fixed length (the distance between the dots in a pair). For the detection of a global orientation structure, local dipole orientations were arranged in a concentric Glass pattern. Thresholds as a function of a stimulus area were determined by measuring the minimum proportion of dipoles among random-dot noise (signal-to-noise ratio) required for the detection of dipoles (features), as well as for the detection of an orientation structure. Thresholds for feature detection were significantly higher than those for the detection of the global structure--regardless of the stimulus size. Spatial integration, however, did not differ between the two tasks: the exponents of the power functions fitted to data for six observers were -0.48 +/- 0.07 for random dipole orientations and -0.62 +/- 0.1 for Glass patterns.