We are familiar with large highly-ordered single crystals, such as those used in the semiconductor industry. There the crystalline ordering in the silicon crystals extends for billions and more lattice spacings. As a result, X-ray diffraction (XRD) patterns of these crystals have sharp Bragg peaks. We are also familiar with liquids, where the molecules have a local structure that is very different to that in a crystal. In addition, in liquids there is only ordering over a few molecular diameters, and so they do not have Bragg peaks. But there are materials, where the situation is much less clear. Many of these materials are referred to as amorphous.For example, the structure of amorphous silicon is controversial. There is debate 1-3 over whether it possesses some crystalline ordering of the sort variously referred to as paracrystalline order, nanocrystalline order or medium-range crystalline order, or whether it has no crystalline ordering. We do not know if locally, silicon atoms are arranged in the structure of a liquid, or whether they are in tiny crystals only a few silicon atoms across. There are other systems at the borderline between liquid and crystal. The mineral ferrihydrite can be classified as crystalline (this is called six-line ferrihydrite), or as amorphous (this is called two-line ferrihydrite). [4][5][6] The structures of Amorphous Calcium Carbonate (ACC), 7-16 Amorphous Calcium Phosphate (ACP), 17 and Amorphous Calcium Sulphate (ACS) [18][19][20] are also poorly understood. The properties of ACC clearly depend on how it is prepared, 9,21-24 and Gebauer et al. 23 refer to 'protovaterite' and 'proto-calcite' forms of ACC, which * To whom correspondence should be addressed although amorphous have NMR and EXAFS spectra that have (broadened) features in common with vaterite, and with calcite, respectively. Vaterite and calcite are two of the crystal polymorphs of calcium carbonate.Experimental data such as XRD can clearly distinguish between large single crystals and liquids, but as we have known since Scherrer's work a hundred years ago, 25 as the size of crystallites decreases, the width of the Bragg peaks in the structure factor S(k) increases. Finite size effects give a peak width ∆k ≈ k/n, for a crystallite n lattice spacings across. For n 10, the peak is so broad that it is essentially as broad as features in the S(k)'s of liquids, and so XRD is not be able to distinguish between a liquid or glass, and a state made of very small crystallites. In computer simulations we can circumvent the problem this causes in identifying a state, as there we know the exact positions of all molecules, in real space, and so have much more information than is contained in an XRD pattern.Here we use computer simulation to study crystallisation in microscopic detail, in a simple model: the Gaussian Core Model (GCM). 26 In the GCM, molecules interact via the spherically symmetric potential v(r) = ε exp[−r 2 /σ 2 ], where the parameters ε and σ have dimensions of energy and length, respectively. This is a simple mod...