The most widely used methods for heliostat field optimization assume a particular pattern and then optimize the parameters that define the pattern to obtain a specified objective such as optimum optical efficiency or maximum power output. It has been demonstrated that these optimized patterns are not necessarily optimal; improvements are possible. Allowing heliostat field placement independent of an a priori pattern may therefore result in improved performance. While initial placement of heliostats in a fixed pattern may reduce computational burden, this paper demonstrates that an optimal field can be designed using a gradient-based optimization method starting from a random initial field and resulting in a field performing better than optimized patterns. The greatest challenge is computational expense. However, a simplified field analysis tool used as an objective function in a suitable optimization procedure sufficiently reduces the computational expense. An algorithm for very large scale optimal design, denoted SAOi, is applied. The algorithm is based on sequential approximate optimization and exploits some of the advantages of quadratic approximations while minimizing storage requirements. This algorithm proves attractive for the present problem where the design variables are large and the constraints outnumber the design variables. The procedure is applied to redesign the PS10 field with an improvement of 1.2% in annual intercepted energy. The research presented herein shows how heliostat placement can be done with each heliostat location as a separate design variable resulting in wholly optimal field placements.