Color separation is an essential step of the data processing in the four-dy
e fluorescence detection strategy used in automated DNA sequencing. In this
paper, we propose a model to describe the crosstalk phenomenon, and show h
ow the assumptions of the model are supported by experimental data. The cro
sstalk matrix is estimated via a reparameterization based on a mapping betw
een the distribution of fluorescence intensities and that of dye concentrat
ions. An iterative algorithm is designed to implement the estimation. To ev
aluate the color-correction quality of a crosstalk matrix, we propose a qua
ntitative measure based on the distribution of the color-corrected data. We
illustrate this method by applying it to a sequencing trace of slab gel el
ectrophoresis obtained at the Human Genome Center at Lawrence Berkeley Nati
onal Laboratory, and that of capillary electrophoresis provided by the Depa
rtment of Chemistry at UC, Berkeley. The accuracy of this method is also as
sessed by the bootstrap method.