Using examples of the square-and triangular-lattice Heisenberg models we demonstrate that the density matrix renormalization group method (DMRG) can be effectively used to study magnetic ordering in twodimensional lattice spin models. We show that local quantities in DMRG calculations, such as the on-site magnetization M , should be extrapolated with the truncation error, not with its square root, as previously assumed. We also introduce convenient sequences of clusters, using cylindrical boundary conditions and pinning magnetic fields, which provide for rapidly converging finite-size scaling. This scaling behavior on our clusters is clarified using finite-size analysis of the effective σ-model and finite-size spin-wave theory. The resulting greatly improved extrapolations allow us to determine the thermodynamic limit of M for the square lattice with an error comparable to quantum Monte Carlo. For the triangular lattice, we verify the existence of three-sublattice magnetic order, and estimate the order parameter to be M = 0.205(15).PACS numbers: 74.45.+c,74.50.+r,71.10.Pm Two-dimensional (2D) quantum lattice systems studied in condensed matter physics can be divided into two types: those with a sign problem in quantum Monte Carlo (QMC), and those without one. This is because recent developments in QMC [1,2,3] have enabled remarkably accurate large-scale studies of the latter systems, such as the square-lattice Heisenberg model (SLHM) [4]. In contrast, the former systems, such as the triangular lattice Heisenberg model (TLHM) and other models with geometric frustration, are often the subject of controversy even regarding questions of what type of order, if any, is present. For the TLHM, it is only recently that the rough agreement between several theoretical[5] and numerical [6,7,8] methods has made a convincing case that the model has three-sublattice, non-collinear 120• order.The density matrix renormalization group[9] (DMRG) is not subject to the sign problem, it has an error which can be systematically decreased by keeping more states, and even with modest computational effort it is extremely accurate for one dimensional and ladder systems. For 2D systems, the computational effort grows exponentially with the width. Ameliorating this effect is the very systematic behavior of the DMRG results versus the number of states kept, enabling the use of extrapolations to improve the accuracy. The extrapolation of the energy versus the truncation error ε (also known as the discarded weight) to the limit ε → 0 often can improve the accuracy of the energy by nearly an order of magnitude. For observables other than the energy, extrapolation has been more problematic and is much less used.In this Letter we show that the difficulty in extrapolating local measurements A is due to the incorrect assumption that the error ∆A ∼ ε 1/2 . In fact, the simplest way to measure local quantities within DMRG makes ∆A analytic in ε. The resulting improved extrapolations greatly improve one's ability to measure order parameters in two ...