It is shown that mean flow similarity based on a velocity scale related to the maximum shear stress (SchofieldPerry similarity) can accurately describe detached two-dimensional turbulent boundary layers provided the origin of the similarity is relocated on the zero velocity streamline in the detached flow. In support of this, data from several different experiments are analyzed and presented. The failure of the standard logarithmic law to accurately describe flow close to the wall in a separating layer is discussed. It is argued that the failure of the logarithmic law is related to the nature of turbulent separation, which is not an event but a process in which the proportion of intermittent flow reversal near the wall gradually increases with distance downstream and therefore, the mean velocities measured within the detachment region contain a proportion of reversed flow which does not follow standard wall similarity. Experimental evidence supports another proposition that a twodimensional turbulent boundary layer detaches with a universal mean profile shape which is accurately described by Schofield-Per ry similarity. After detachment, the outer separated layer shows only small development, so that reattachment occurs with only a slightly different universal mean profile shape.