In the laboratory, thresholds for stereoscopic depth perception are usually determined by asking observers to discriminate between a stimulus with a given depth offset and its mirror image. Threshold is most often defined as the disparity difference that yields 75% or 83% correct responses. Disparities used for clinical tests of stereopsis are much higher. Here it is argued that, among other factors, this is because of the fact that clinical tests usually require the detection of a depth difference (offset versus no offset), rather than the discrimination between two directions of depth difference (in front versus behind). From a formal comparison of the two tasks, the data show that discrimination, or classification is easier by at least a factor of 2 than detection. The contribution of variations of the threshold criterion and learning to the differences between stereoacuity as measured in laboratory and clinic is also discussed. These differences are relevant to the design of tests for clinical use.