It is generally agreed that, above the level of the 30 nm fibre, eukaryotic chromatin is constrained into loops, but there is disagreement about the nature of the substructure that serves to anchor loops and the DNA sequences that act as the attachment sites. This problem may stem from the very different methods that all purport to separate loop and attached DNAs. We have tested ideas about how the genome is arranged into loops by analysing the average loop size over different cytologically resolvable regions of human chromosomes using fluorescence in situ hybridisation with loop and attached DNA fractions. Variations in average loop size, along and between chromosomes, measurable at this level of resolution were small but significant and were dependent on the extraction method. This emphasises the fundamental differences between the nuclear substructure probed by different protocols. DNA attached to the nuclear 'scaffold' or 'matrix' hybridises preferentially to gene-poor regions of the genome (G-bands). Conversely, fractions attached to the nuclear 'skeleton' hybridise preferentially to gene-rich R-bands and sites of high levels of transcription. The inactive X chromosome has a deficit of associations with the nuclear skeleton but not with the matrix or scaffold. A large excess of attached sequences is found at some sites of constitutive heterochromatin, but not at centromeres.