Non-adjacent dependencies (NADs) refer to dependencies between items that are not adjacent in a sequence. Peña et al. (2002) discovered adult participants could learn the NADs of syllables in an artificial language when there were 25ms pauses before and after the NADs. Studies using videos of human body movements showed similar learning outcomes (Endress & Wood, 2011). However, participants failed to learn the NADs with respect to non-linguistic acoustic stimuli, such as tones or noises (Gebhart, Newport, & Aslin, 2009). Four experiments in this study examined the constraints on learning the NADs of visual stimuli. We propose that acquisition of the NADs requires the sequences be packed into a coherent unit, and the motor system provides the require packaging for stimuli that can be mapped onto motor representation. Implications on the acquisition of syllable NADs are discussed.