Collagen XV is a million-dalton protein with a structural role in skeletal muscle and capillaries. As with all collagens, studies of its function are hindered by the absence of good structural data: collagens are triple-helical, non-crystallizable, multidomain proteins with extensive post-translational modification that are refractory to analysis by high-resolution structural techniques. For collagen XV, this situation is compounded by the fact that it is also a proteoglycan. In this issue of the Biochemical Journal, Myers and her colleagues use rotary shadowing electron microscopy to obtain images of purified collagen XV molecules that are sufficiently detailed to show the three-lobed structure of the N-terminus and individual glycosaminoglycan side chains. Individual molecules appear as knotted strands resembling a pretzel (a pastry snack folded in a unique figure-of-eight), which contrasts with our conventional image of collagen molecules as semi-rigid rods. Importantly, collagen XV multimerizes into cruciform structures in which simpler forms have two to four molecules per complex. Immunoelectron microscopy revealed knotted collagen XV complexes bridging collagen fibrils adjacent to basement membrane. These accomplishments are made all the more impressive by the fact that collagen XV was purified from human umbilical cord, in which the protein is represented at only (1−2) × 10 −4 % of dry weight! Key words: basement membrane, collagen, electron tomography, endostatin, extracellular matrix, fibril, rotary shadowing.Understanding the structure and assembly of collagens has occupied the attention of matrix biologists and structural biologists for many years. X-ray fibre diffraction studies in the 1950s and recent X-ray crystallography of collagen-like peptides have provided information about the collagen triple helix at atomic resolution (for reviews, see [1,2]). However, most collagens function as part of supramolecular assemblies, which are refractory to study by conventional high-resolution techniques. Consequently, important questions about the structure of native collagens and how they assemble in vivo remain unanswered.Studies in the 1960s-1980s used electron microscopy to record detailed images of type I collagen (the major collagen, found in skin and tendon) and the 67 nm D-periodic fibrils that are formed by self-assembly of type I collagen molecules (for a review, see [3]). In subsequent years other collagens were identified that assemble into extended networks (e.g. collagens IV, VIII and X), anchoring filaments (collagen VII), beaded narrowdiameter microfibrils (e.g. collagen VI), and FACIT (fibril-associated collagens with interrupted triple helices) collagens that bind to the surfaces of collagen fibrils and provide binding sites for ECM (extracellular matrix) macromolecules (for a review, see [4]). We could be forgiven for thinking that we had seen the complete array of supramolecular assembles that collagen could form. Then along came the paper by Myers and colleagues in this issue of the Bioche...