In this paper, we consider a massive multiple-input multiple-output (MIMO) system for urban macro cellular networks. A new technique named vertical spatial filtering (VSF) is proposed to achieve low radio frequency (RF) complexity by exploiting the spatial beam redundancy in elevation domain. Specifically, VSF consists of two parts: beam mapping and beam selection. We first adopt beam mapping to transform the received signals from antenna space into beam space. Then, we use beam selection to pick out part of the RF chains so as to realize the reduction of RF complexity. A three-dimensional channel model is adopted to evaluate the system performance. Two strategies for beam selection are proposed and numerical results are presented. The results show that the reduced system with VSF has the performance comparable to the complete system, while only a small amount of RF chains is required in vertical dimension of the two-dimensional base station array.