Procedures for the seismic design of new buildings to achieve a minimum acceptable performance are well established. However, it is concluded in this paper that the actual performance of structures will not be predicted with any certainty using the same procedures, given the uncertainties associated with the assumptions that must be made. Sole reliance on design procedures to predict performance is likely to lead to unnecessary conservatism and this must be taken into account when setting standards for design. It is also concluded that there is currently insufficient data available in a suitable form to enable the correlation between analysis and actual performance to be assessed in any way other than by judgement. The prediction of seismic performance remains very much an art.