[en] This thesis documents the development and application of a computational algorithm for Large-Eddy Simulation. Unusually, the method adopts a fully collocated variable storage arrangement and is applicable to complex, non-rectilinear geometries. A Reynolds-averaged Navier-Stokes algorithm has formed the starting point of the development, but has been modified substantially: the spatial approximation of convection is effected by an energy-conserving central-differencing scheme; a second-order time-marching Adams-Bashforth scheme has been introduced; the pressure field is determined by solving the pressure-Poisson equation; this equation is solved either by use of preconditioned Conjugate-Gradient methods or with the Generalised Minimum Residual method; two types of sub-grid scale models have been introduced and examined. The algorithm has been validated by reference to a hierarchy of unsteady flows of increasing complexity starting with unsteady lid-driven cavity flows and ending with 3-D turbulent vortex shedding behind a square prism. In the latter case, for which extensive experimental data are available, special emphasis has been put on examining the dependence of the results on mesh density, near-wall treatment and the nature of the sub-grid-scale model, one of which is an advanced dynamic model. The LES scheme is shown to return time-average and phase-averaged results which agree well with experimental data and which support the view that LES is a promising approach for unsteady flows dominated by large periodic structures. (author)