Two methods are currently used to calculate the expected loss for existing buildings for a time period of interest. The first method simply multiplies the annual expected loss by the time period. The second method uses the exceedance probability for that time period to calculate the expected loss. Results from these methods can be significantly different depending on the length of the time period and the loss function of the building class. Generally speaking, the first method overestimates the risk while the second method underestimates it. In this paper, it is shown theoretically that differences between the two methods are due to differences in how multi-occurrences of earthquakes during the period are incorporated. Neither method is quite consistent with the physical process of earthquake occurrence. A new method which is more consistent with the occurrence process is suggested. Numerican examples are given to illustrate the accuracy of the new method, and how it compares with existing approaches. The new method has been incorporated into a personal computer analysis program called EXLOSS, which is available from the John A. Blume Earthquake Engineering Center of Stanford University.