Most methods for deriving Q from surface-seismic data depend on the spectral content of the reflection. The spectrum of the reflected wave may be affected by the presence of thin beds in the formation, which makes Q estimates less reliable. We incorporate a method for correcting the reflected spectrum to remove local thin-bed effects into the Q-versus-offset (QVO) method for determining attenuation from seismic-reflection data. By dividing the observed spectrum by the local spectrum of the known reflectivity sequence from a nearby well log, we obtain a spectrum more closely resembling that which would be produced by a single primary reflector. This operation, equivalent to deconvolution in the time domain, is demonstrated to be successful using synthetic data. As a test case, we also apply the correction method to QVO with a real seismic line over a south Florida site containing many thin sandstone and carbonate beds. When corrected spectra are used, there is significantly less variance in the estimated Q values, and fewer unphysical negative Q values are obtained. Based on this method, it appears that sediments at the Florida site have a Q near 33 that is roughly constant from 170- to 600-m depth over the length of the line.