We establish methods for quantum state tomography based on compressed sensing. These methods are specialized for quantum states that are fairly pure, and they offer a significant performance improvement on large quantum systems. In particular, they are able to reconstruct an unknown density matrix of dimension d and rank r using O(rd log 2 d) measurement settings, compared to standard methods that require d 2 settings. Our methods have several features that make them amenable to experimental implementation: they require only simple Pauli measurements, use fast convex optimization, are stable against noise, and can be applied to states that are only approximately low-rank. The acquired data can be used to certify that the state is indeed close to pure, so no a priori assumptions are needed. We present both theoretical bounds and numerical simulations.The tasks of reconstructing the quantum states and processes produced by physical systems -known respectively as quantum state and process tomography [1] -are of increasing importance in physics and especially in quantum information science. Tomography has been used to characterize the quantum state of trapped ions [2] and an optical entangling gate [3] among many other implementations. But a fundamental difficulty in performing tomography on many-body systems is the exponential growth in the state space dimension. For example, to get a maximum-likelihood estimate of a quantum state of 8 ions, Ref.[2] required hundreds of thousands of measurements and weeks of post-processing.Still, one might hope to overcome this obstacle, because the vast majority of quantum states are not of physical interest. Rather, one is often interested in states with special properties: pure states, states with particular symmetries, ground states of local Hamiltonians, etc., and tomography might be more efficient in such special cases [4].In particular, consider pure or nearly pure quantum states, i.e., states with low entropy. More precisely, consider a quantum state that is essentially supported on an r-dimensional space, meaning the density matrix is close (in a given norm) to a matrix of rank r, where r is small. Such states arise in very common physical settings, e.g. a pure state subject to a local noise process [20].A standard implementation of tomography [5,6] would use d 2 or more measurement settings, where d = 2 n for an nqubit system. But a simple parameter counting argument suggests that O(rd) settings could possibly suffice -a significant improvement. However, it is not clear how to achieve this performance in practice, i.e., how to choose these measurements, or how to efficiently reconstruct the density matrix. For instance, the problem of finding a minimum-rank matrix subject to linear constraints is NP-hard in general [7].In addition to a reduction in experimental complexity, one might hope that a post-processing algorithm which takes as input only O(rd) ≪ d 2 numbers could be tuned to run considerably faster than standard methods. Since the output of the procedure is a low...