This presentation gives an overview of sampling techniques used in forest inventories. The starting point is the reformulation of the Horwitz-Thompson estimator as a local density: estimating the discrete sum over a nite population of trees is transformed into the estimation of the integral of a function (in nite population model or Monte Carlo approach). The highlight of this introductory part is the famous angle count method of Bitterlich, the only exact PPS sampling scheme directly implementable in practice. The basic concepts are then generalized to two-stage Poisson sampling, cluster sampling and two-phase sampling, where the use of remote sensing data plays a key role. A brief comparison with the model dependent Kriging techniques is also given. The optimization of sampling schemes by considering the anticipated variance under local Poisson models for the spatial distribution of the trees is sketched. An inventory at the enterprize level, simulations and data from the Swiss National Inventory illustrate the theoretical developments.