The imprinted Snurf-Snrpn domain, also referred to as the Prader-Willi syndrome region, contains two ∼100-200 kb arrays of repeated small nucleolar (sno)RNAs processed from introns of long, paternally expressed non-protein-coding RNAs whose biogenesis and functions are poorly understood. We provide evidence that C/D snoRNAs do not derive from a single transcript as previously envisaged, but rather from (at least) two independent transcription units. We show that spliced snoRNA host-gene transcripts accumulate near their transcription sites as structurally constrained RNA species that are prevented from diffusing, as well as multiple stable nucleoplasmic RNA foci dispersed in the entire nucleus but not in the nucleolus. Chromatin structure at these repeated arrays displays an outstanding parent-of-origin-specific higher-order organization: the transcriptionally active allele is revealed as extended DNA FISH signals whereas the genetically identical, silent allele is visualized as singlet DNA FISH signals. A similar allele-specific chromatin organization is documented for snoRNA gene arrays at the imprinted Dlk1-Dio3 domain. Our findings have repercussions for understanding the spatial organization of gene expression and the intra-nuclear fate of non-coding RNAs in the context of nuclear architecture.