Many animal species can detect dependencies between adjacent visual or auditory items in a string. Compared with adjacent dependencies, detecting nonadjacent dependencies, as present in linguistic constructions, is more challenging as this requires detecting a relation between items irrespective of the number and nature of the intervening items. There is limited evidence that nonhuman animals can detect such dependencies. An animal group in which such abilities might be expected is songbirds, which have learned songs consisting of a series of vocal elements given in specific sequences. So far no songbird (or other bird species) has been tested for its ability to detect nonadjacent dependencies. We examined whether zebra finches can detect the dependencies between items at the edges of artificially arranged strings of song elements. Zebra finches were trained to discriminate 2 sets of dependent song elements that always appeared in the same order (A and B; C and D), from other element combinations (AD, AC, BD, CB, CA, DB). The element combinations were separated by intervening (I) elements. Subsequent tests revealed that the finches could generalize the learned dependencies over different numbers and types of intervening items. Our findings show that the ability for detecting nonadjacent dependencies is not limited to humans or primates, and lend support to theories that suggest that nonadjacent dependencies can be learned by a nonlinguistic associative learning process.