Dolphin structures can effectively mitigate the hazards of ship collisions posing a significant risk of severe damage to long-span bridges. Several studies have quantitatively assessed the ship collision risks of long-span bridges surrounded by dolphin structures. However, the decision-making procedure of dolphins based on the risk analysis has not been investigated with a systematic treatment of uncertainties. To this end, this study proposes an optimization method for the decision-making of dolphin structures using system-reliability-based disaster resilience analysis for pylon failure scenarios. First, we introduce the decision optimization method named “Reliability-Redundancy-Recoverability-based Decision Optimization (R3-DO)” defined by the objective function representing the recoverability and constraints on the system-level resilience. Next, the primary uncertainties in ship-collision events and the design parameters of dolphin design are identified based on the data from the Incheon Grand Bridge of South Korea. The ship collisions with dolphins and the pylon are modeled through surrogate models of finite element simulations to facilitate repetitive simulation-based reliability analyses and the following gradient-based optimization. The results demonstrate that the proposed R3-DO effectively finds the optimal design of the dolphin structures considering uncertainties of ship-collision hazards and the system-level resilience of long-span bridges.