When subjects learn to associate two sample durations with two comparison keys, do they learn to associate the keys with the short and long samples (relational hypothesis), or with the specific sample durations (absolute hypothesis)? We exposed 16 pigeons to an ABA design in which phases A and B corresponded to tasks using samples of 1 s and 4 s, or 4 s and 16 s. Across phases, we varied the mapping between the samples and the keys. For group Relative, short and long samples were always associated with the same keys (e.g., Phase A: "1s [right arrow] Left, 4s [right arrow] Right"; Phase B: "4s [right arrow] Left, 16s [right arrow] Right"); for group Absolute, the 4-s sample was associated always with the same key (e.g., Phase A: "1s [right arrow] Left, 4s [right arrow] Right"; Phase B: "16s [right arrow] Left, 4s [right arrow] Right"). If temporal control is relational, group Relative should learn the new task faster than group Absolute, but if temporal control is absolute, the opposite should occur. We compared the results with the predictions of the Learning-to-Time (LeT) model, which accounts for temporal discrimination in terms of absolute stimulus control and stimulus generalization. The acquisition curves of the two groups were generally consistent with LeT and therefore more consistent with the absolute than the relative hypothesis. (Contains 1 footnote, 3 tables, and 11 figures.)