Spatial and temporal properties of illusory contours and amodal completion were investigated using a shape discrimination task. Performance was characterized as accuracy of angular discrimination of the inducing figures ("pacmen") in a two-alternative forced choice (2AFC) paradigm. First, we compared performance when four "pacmen" were organized into Kanizsa-like figures (squares and small deformations of squares) which produced the percept of illusory contours (ICs), with performance obtained with all four "pacmen" facing in the same direction, when no illusory contours were seen. Then, we found that it was possible to interfere with boundary completion and degrade performance with masking lines placed between the inducers of a Kanizsa figure. From these experiments we concluded that performance in the shape discrimination task depended on boundary completion. Next, the dependence of contour-dependent performance on the spatial scale of the figures was examined. Threshold angular discrimination was approximately scale-invariant and subjects were able to integrate visual information across gaps as large as 13 deg of visual angle. Performance in the shape recognition task for illusory and amodally completed figures was also measured. Similar accuracy was obtained either when the boundaries were modally or amodally completed. Finally, we used shape discrimination in conjunction with backward masking to explore the dynamics of boundary completion. Two different phases of the boundary completion process were observed. The first phase was revealed when the inducers were locally masked, and took approximately 117 msec. A second phase lasted an additional 140-200 msec after the inducers were masked.