An application is presented of a recently proposed system identification method for buildings that takes into account the effects of soil-structure interaction and the coupling of the horizontal and rocking motions of the foundation. The method gives the uncoupled structural fixed-base frequency f (sub 1) and rigid-body rocking frequency f (sub R) using data only from two horizontal sensors (at base and roof). The fixed-base frequency is estimated from the wave travel time through the structure, the apparent system frequency from Fourier analysis, and the rocking frequency from a relation between these three. The case study is Millikan Library in Pasadena, California. Results are shown for four earthquakes between 1970 and 2002. The method makes it possible to quantify the degree to which the observed changes (wandering) of its resonant frequencies have been due to changes in the structure alone. The results show that (1) both f (sub 1) and f (sub R) are amplitude dependent, (2) significant permanent reduction of frequency occurred over the years, approximately 22% for f (sub 1) and 11% for the apparent frequency f (sub 1,app) , mostly caused by the San Fernando earthquake of 1971, while (3) the changes of f (sub R) have been amplitude dependent and recoverable. (4) During the San Fernando earthquake, both f (sub 1) and f (sub R) dropped, respectively, by approximately 24% and approximately 18%, resulting in 21% drop of f (sub 1,app) . (5) After this earthquake, the changes in the observed resonant frequencies, which are those of the system, have been due to a much larger degree (4-5 times) to changes of f (sub R) than to changes of f (sub 1) . (6) The small permanent changes in f (sub 1) that appear to have occurred after the San Fernando earthquake cannot be deciphered with certainty because of the small number of earthquake records available for this analysis since 1971. Records from the period 1988 to 2002, when released, can be used in future to refine and verify these trends.