The base station (BS) density configuration is a key factor to improve energy efficiency (EE) performances. In this paper, BS density configurations for achieving the optimal EE performance in two-tier cellular networks are analyzed, where the Poisson point process (PPP) is used to model the BS spatial distribution. To make the EE performances trackable and analyzable, the BS density optimization in each tier is transformed into an equivalent problem that jointly optimizes "the sum and the ratio of BS densities". This equivalent optimization problem is not necessarily convex, while its monotonicity with different power consumptions of BSs is analyzable. Considering the optimal BS density configuration is not unique and the closed-form solution for achieving the best EE performance is extremely difficult to be derived, a dynamic gradient based iterative algorithm by solving the quadratic functions is proposed. Furthermore, the quantitative analysis of EE performances based on the data fitting method shows that the approximately linear relationship between the optimal BS density and the user density holds only under specific conditions. Simulation results have demonstrated the effectiveness of the proposed algorithm and verified relevant conclusions.