Motivated by the intriguing report, in some frustrated quantum antiferromagnets, of magnetization plateaus whose simple collinear structure is not stabilized by an external magnetic field in the classical limit, we develop a semiclassical method to estimate the zero-point energy of collinear configurations even when they do not correspond to a local minimum of the classical energy. For the spin-1/2 frustrated square-lattice antiferromagnet, this approach leads to the stabilization of a large 1/2 plateau with "up-up-up-down" structure for J 2 /J 1 > 1/2, in agreement with exact diagonalization results, while for the spin-1/2 anisotropic triangular antiferromagnet, it predicts that the 1/3 plateau with "up-up-down" structure is stable far from the isotropic point, in agreement with the properties of Cs 2 CuBr 4 .Introduction.-Frustration is responsible for the emergence of several remarkable properties in quantum magnets, ranging from rather exotic types of order such as quadrupolar or nematic order to resonating valence bond or algebraic spin liquids [1]. In the presence of an external field, frustration is also known to be at the origin of several types of accidents in the magnetization curve, including kinks, jumps and plateaus. Of all these remarkable features, magnetization plateaus at rational value of the magnetization are probably the best documented ones experimentally, and their theory is likewise quite advanced. Following the terminology of Hida and Affleck [2], two kinds of plateaus have been identified [3]: 'classical' plateaus [4][5][6], whose structure has a simple classical analog with spins up or down along the external field, and 'quantum' plateaus [7][8][9][10][11], which have no classical analog and correspond to a Wigner crystal of triplets in a sea of singlets. In the case of quantum plateaus, the mechanism is clear: frustration reduces the kinetic energy of triplets, resulting in a crystallization at commensurate densities. The main open problem is to be predictive for high commensurability plateaus since it requires a precise knowledge of the long-range part of the triplet-triplet interaction.By contrast, and somehow surprisingly, the theory of classical plateaus is not complete yet. The paradigmatic example of a classical plateau is the 1/3 magnetization plateau of the Heisenberg antiferromagnet on a triangular lattice, studied by Chubukov and Golosov [5] in the context of a 1/S expansion. In this system the three sublattice up-up-down (uud) structure appears classically at H = H sat /3, and quantum fluctuations stabilize this uud state in a finite field range around H sat /3, leading to the 1/3 plateau. The basic qualitative idea in the spirit of the order by disorder is that collinear configurations often have a softer spectrum and, hence, a smaller zeropoint energy [12,13]. The prediction of the 1/3 plateau has been confirmed by exact diagonalization of finite clusters for S = 1/2 and 1 [14], and the theory of Chubukov and Golosov can be extended to all cases where a collinear state is ...