Collagen is most abundant in animal tissues as very long fibrils with a characteristic axial periodic structure. The fibrils provide the major biomechanical scaffold for cell attachment and anchorage of macromolecules, allowing the shape and form of tissues to be defined and maintained. How the fibrils are formed from their monomeric precursors is the primary concern of this review. Collagen fibril formation is basically a self-assembly process (i.e. one which is to a large extent determined by the intrinsic properties of the collagen molecules themselves) but it is also sensitive to cell-mediated regulation, particularly in young or healing tissues. Recent attention has been focused on ‘early fibrils’ or ‘fibril segments’ of ~10 μm in length which appear to be intermediates in the formation of mature fibrils that can grow to be hundreds of micrometres in length. Data from several laboratories indicate that these early fibrils can be unipolar (with all molecules pointing in the same direction) or bipolar (in which the orientation of collagen molecules reverses at a single location along the fibril). The occurrence of such early fibrils has major implications for tissue morphogenesis and repair. In this article we review the current understanding of the origin of unipolar and bipolar fibrils, and how mature fibrils are assembled from early fibrils. We include preliminary evidence from invertebrates which suggests that the principles for bipolar fibril assembly were established at least 500 million years ago.