Abstract A review of the techniques of computer modelling of amorphous magnetic materials is presented. The magnetic materials discussed fall mainly into three categories, the ferromagnetic transition-metal metalloid alloys, the ‘ferrimagnetic’ transition-metal rare-earth alloys, and the speromagnetic amorphous magnetic insulators. Although all have primarily been computer modelled using simulations based on a common structural concept — that of the dense random packing of hard spheres (DRPHS) — details of the various modelling techniques are different for each. After introducing the basic DRPHS technique for the simplest possible case involving spheres of one size, and a discussion of its relevance in describing the rather unstable amorphous elemental magnetic metals, the importance of a subsequent relaxation (or energy minimization) of the hard-sphere aggregates into ‘soft’-sphere equivalents is stressed. Comparison with experiment occurs primarily via the diffraction measurement of radial distribution functions. The review then progresses to an analysis of binary hard-sphere assemblies corresponding to stable amorphous magnetic alloys, and sets out convincing evidence for the importance chemical bonding in the metalloid systems. It focuses, in particular, on the ways in which such bonding forces can be accommodated within computerized building schemes both at the hard-sphere and soft-sphere levels of the development. Although chemical constraints are certainly less severe in the amorphous transition-metal rare-earth alloys, evidence for their possible importance in this context is also covered. The review concludes with an account of the modelling of amorphous magnetic insulators, using the ternary system of amorphous yttrium iron garnet as an example. The importance of including strong Coulomb forces between highly ionized constituents is demonstrated, and a method of incorporating them into the DRPHS computer building scheme is described. Speculations concerning the possible breakdown of DRPHS methods and the possible future relevance of random-network modelling for other ‘less-densely-packed’ magnetic insulator glasses are set out.