This work presents a two-phase sampling approach to address reliability-based optimization problems in structural engineering. The constrained optimization problem is converted into a sampling problem, which is then solved using Markov chain Monte Carlo methods. First, an exploration phase generates uniformly distributed feasible designs. Thereafter, an exploitation phase is carried out to obtain a set of close-to-optimal designs. The approach is general in the sense that it is not limited to a particular type of system behavior and, in addition, it can handle constrained and unconstrained formulations as well as discrete–continuous design spaces. Three numerical examples involving structural dynamical systems under stochastic excitation are presented to illustrate the capabilities of the approach.