Structural damage and collapse of bridges due to fire have increased in recent years, emphasising the need for appropriate fire-safety-oriented design approaches. Current methods of fire-safe design often consist of a simple extrapolation of rules used originally for buildings and do not provide bridge-specific guidance as for other hazards. Available design methodologies (deterministic and/or probabilistic) for extreme loads start by characterising the hazard at the site. Next, they involve structural response analysis to estimate hazard consequences in terms of damage or loss metrics of interest. These metrics are eventually appraised to verify whether various performance objectives are achieved. While the same approach has been applied to fire (as an extreme load), the results’ appropriateness is unclear. Dynamic coupling between structures and fire exists, where the fire affects a structure, but the characteristics of the structure also affect the fire. This makes the workflow mentioned above unsuitable for delivering optimised solutions. Indeed, the structural design choices define the fire scenarios that could potentially affect the structure over its lifespan. This paper proposes a Maximum Allowable Consequence (MAC) approach to the fire safety design of bridges that considers fire scenarios as additional design variables and delivers them as outputs. A single-span bridge subject to a car fire is considered to illustrate the proposed MAC approach, and the following performance metrics are selected: 1) the time 𝑡𝑢 to reach unsatisfactory conditions (i.e., the bridge’s collapse); 2) the probability of that time being lower than a MAC threshold of 20 min. The design variables are the clearance and girder height. Numerical optimisation is applied to calculate fire scenario features that minimise 𝑡𝑢. Then, the effect of input uncertainties in steel material properties on the estimated fire consequences is investigated through Monte Carlo sampling. Finally, these metrics inform the selection of an optimal design configuration.