Recent studies showing learners can induce phrase structure from distributional patterns (Thompson & Newport, 2007; Saffran, 2001) suggest that phrase structure need not be innate. Here, we ask if this learning ability is restricted to language. Specifically, we ask if phrase structure can be induced from non-linguistic visual arrays and further, whether learning is assisted by abstract category information. In an artificial visual grammar paradigm where co-occurrence relationships exist between categories of objects rather than individual items, participants preferred phrase-relevant pairs over frequency-matched non-phrase pairs. Additionally, participants generalized phrasal relationships to novel pairs, but only in the cued condition. Taken together these results show that learners can acquire phrase structure in a non-linguistic system, and that cues improve learning.