A frequency domain deconvolution method to extract past input fluxes of particle-associated pollutants from their sedimentary records is developed. The method is based on the assumption of a mixing layer below which no further mixing takes place. From the latter historical layer a filtered time record is obtained. Also, from known steady state sedimentation parameters, a response to a unit impulse input is calculated by forward finite difference programs. The unknown input flux is then generated by taking Fourier transforms of the two time series, forming the ratio of the transforms, and applying the inverse frequency transform. The method has been tested for internal consistency and applied to Pb, Zn, and Cd in several sediment cores from Lake Michigan. The results are consistent with those obtained previously with a time domain method. However, at the present stage of development, the time resolution in the frequency method (≈1 year) is considerably better than in a general time domain method (≈15 years). The effects of deconvolution are to sharpen peaks, to generate peaks where none were present in the unprocessed record, and to shift the onset time of pollution towards more recent times. In most cases, there is a remarkable agreement between reconstructed input records and alternative source functions. For example, maxima around 1969 for Pb and Cd in deconvolved input records are confirmed by the atmospheric loading data for the Great Lakes region (Pb) and U.S. consumption figures (Cd). ¿ American Geophysical Union 1989