Structural determination of individual protein domains isolated from multidomain proteins is a common approach in the post-genomic era. Novel and thus uncharacterized domains liberated from intact proteins often self-associate due to incorrectly defined domain boundaries. Self-association results in missing signals, poor signal dispersion and a low signal-to-noise ratio in (1)H-(15)N HSQC spectra. We have found that a putative, non-canonical coiled coil region close to a domain boundary can cause transient hydrophobic self-association and monomer-dimer equilibrium in solution. Here we propose a rational method to predict putative coiled coil regions adjacent to the globular core domain using the program COILS. Except for the amino acid sequence, no preexisting knowledge concerning the domain is required. A small number of mutant proteins with a minimized coiled coil region have been rationally designed and tested. The engineered domains exhibit decreased self-association as assessed by (1)H-(15)N HSQC spectra with improved peak dispersion and sharper cross peaks. Two successful examples of isolating novel N-terminal domains from AAA-ATPases are demonstrated. Our method is useful for the experimental determination of domain boundaries suited for structural genomics studies.