In six experiments using the speeded classification paradigm, we provide evidence that the ostensibly "separable" dimensions of size and orientation can produce patterns of either separability or asymmetric configurality, depending on the spatial arrangement of the stimuli. In all experiments, subjects classified large or small circles containing a single line in one of two possible orientations. When the line touched the circle's perimeter, thereby defining the diameter of the circle (Experiments 1-4), asymmetric configurality obtained: Variations in size interfered with classification by orientation, but variations in orientation did not interfere with classification by size, and redundancy gain was weak or absent. When the lines fell completely within (i.e., did not touch) the circles (Experiments 5 and 6), the results were consistent with separability: There was neither redundancy gain nor interference. Taken together, the results add to the growing body of evidence that classification of specific dimensional pairs as separable or integral may be less feasible than identifying the more general conditions that increase or decrease the psychological salience of dimensional structures and facilitate or interfere with selection of optimal processing strategies.