Single Particle Analysis by Electron Microscopy aims at producing a three-dimensional model of a biological macromolecule using projection images acquired with an electron microscope. The task boils down to solving the inverse problem of estimating the three-dimensional structure from thousands of two-dimensional projections of it. The reconstruction process is iterative and needs an initial guess of the structure that can be obtained by several methods from the acquired data itself. The algorithm presented in this article eliminates the need of manually choose one of these low-quality volumes. Instead, it considers the whole population of initial volumes along with the acquired data and allows the whole population to evolve according to the dynamics given by swarm optimization. We show that this strategy successfully finds good initial estimates without the need of user intervention.