Seismic base isolation is a well-known seismic protection system that is used to protect structures from earthquakes. The superior seismic performance of base isolated structures has been proven in both analytical simulations and in real earthquakes. However, the use of base isolation in buildings is typically associated with an increase in construction cost. The decision of whether or not to incorporate base isolation in a new building design thus involves the weighting of predicted initial cost increases against potential benefits over the building life cycle. This dissertation is concerned with the quantification of risks and benefits involved with adopting seismic base isolation, as well as the methods by which these risks and benefits are evaluated. The dissertation has three main parts. In the first part, a detailed case study is set out that compares the performance of a base isolated and a conventionally designed building. Particular attention is paid to moat wall pounding and its financial consequences. The base isolated building demonstrates generally superior performance. However, the performance is dependent on the site class, the building ductility and the building’s seismic gap. Pounding against the moat walls degrades the performance of the isolated building and earthquakes that cause pounding contribute significantly to the building’s expected annual loss. The second part of the dissertation reviews the methods by which the buildings are assessed, with a focus on the FEMA P-58 methodology, and proposes some new methods and extensions to available methods. These include the following: (a) the use of Bayesian statistics to estimate mutually exclusive and simultaneous damage state probabilities including allowance for grouping effects; (b) an informative prior for the Straub and Der Kiureghian (2008) method that can be employed to avoid the simulation of fragility curves with failure probabilities that conflict with the analyst’s subjective judgments; (c) a flexible six parameter fragility model that incorporates both presumed aleatory (within-group) and epistemic (between-group) uncertainties; (d) a method of modelling damage state correlations using copulas; (e) the use of the First Order Second Moment (FOSM) reliability method to model the variation of repair costs with the number of damaged components; and (f) an advanced storey-based loss estimation framework which lumps losses into groups at the floor level while accounting for epistemic uncertainties in component fragilities and intercomponent correlations. The effects of epistemic uncertainties in component fragilities and inter-component correlations on floor group outputs are investigated in detail by way of an illustrative example. The third part of the dissertation applies the new methods in a robust cost-benefit analysis that considers both presumed aleatory and epistemic uncertainties. A framework is proposed for consistent probabilistic performance comparison between base isolated and fixed base structures with dissimilar fundamental periods. The framework is suited for assessing the performance base isolated structures in which moat wall pounding represents a significant source of risk. The method is used to identify the range and likelihood of different net present value outcomes in a set of case study buildings. Epistemic uncertainties are considered in the seismic hazard, the fragility function parameters and the mutually exclusive and simultaneous damage state probabilities. Uncertainty regarding the discount rate, the additional construction cost required to install the base isolation system and the time period are also considered. Of these various sources of uncertainty, uncertainty in the increase in construction cost to incorporate base isolation is found to have the greatest influence on expected annual losses and on likelihoods of positive net present value.