Flanks were added to Titchener's (1901) T-illusion figure to test its susceptibility to context stimuli. The addition of a second divided line yielded H-type figures, and the addition of a second undivided line, +-type figures. The lengths of the Ts' undivided lines was expected to be overestimated relative to the lengths of the divided lines, when all lines were about equally long, and the illusion was expected to become smaller when one or two gaps had been introduced between the lines. Results conformed to the predictions. The amount of illusion was larger for the no-gap H than the T, and was almost annihilated with the two-gaps H, with 3 out of 14 observers showing an inverse response bias. The +-type stimuli produced analogous results. Findings are interpreted in terms of the nonequivalence of the endpoints of the stimuli's lines, which are thought to elicit different responses in end-inhibited cortical neurons, thereby affecting length estimates.