A method for deconvolution of the near-uv second-derivative spectra of proteins into their component tryptophan, tyrosine, and phenylalanine spectra is described. In this approach, the second-derivative spectra of tryptophan and tyrosine model compounds are numerically shifted to create a set of reference spectra corresponding to anticipated peak positions in protein environments of different polarity. The relative contributions of these individual standard spectra are varied until the best fit to the experimental protein spectrum is obtained. Separate addition of tryptophan and tyrosine standard spectra, weighted by their contributions as determined in the fitting procedure, yields an accurate representation of the spectra of these residues in proteins. The position of the intersection of these spectra with the wavelength axis is used as a measure of spectral position in ethylene glycol perturbation experiments in which the average solvent accessibility is assessed by relating the observed shifts in the tryptophan and tyrosine spectra to the shifts observed for corresponding model compounds. The phenylalanine peak positions in the set of 16 proteins studied are determined as described previously [H. Mach et al. (1991) Arch. Biochem. Biophys. 287, 33-40]. For all three aromatic residues in proteins, no consistent correlation between absolute spectral band positions and average solvent accessibility is observed, suggesting a significant influence of other local (e.g., electrostatic) effects on near-uv spectra of proteins. The maximum spectral shift observed between solvent-exposed model compounds and side chains entirely buried in apolar protein core was found to be approximately 5 nm for tyrosine, 4 nm for tryptophan, and 2 nm for phenylalanine residues.