The problem of optimizing antenna locations in a radio telescope with a large number of antennas is addressed. An algorithm is developed that first optimizes the probability density function of the antenna positions and this distribution is subsequently sampled. This approach avoids the large number of variables with many antennas. The density function is optimized subject to terrain constraints and the distribution of visibility samples. The optimization is solved by mapping the problem to a phase retrieval problem which is solved using an iterative projection algorithm. © 2010 IEEE.