In this paper, a seismic reliability-based approach is proposed to design inelastic steel moment frame structures isolated by lead-rubber bearing (LRB) systems. An equivalent two-degree-of-freedom system is assumed in which a bilinear behaviour is assigned to both the superstructure and the base. Furthermore, uncertainties associated with the equivalent superstructure mass, stiffness, and yield properties are taken into account by employing proper probability density functions. The proposed design approach is twofold: 1) Reliability curves that return the key design parameters of the inelastic base-isolated structure including: the period of the superstructure, the target base displacement, and the ductility-dependent strength reduction factor for a given target reliability. 2) Regression equations, which estimate the displacement ductility demand of the inelastic superstructure and the optimal design properties of the lead-rubber bearing system including: the total initial stiffness and the total yield force. These regression equations are calibrated against a large set of optimally-designed base-isolated buildings using Genetic Algorithm.