To investigate the sequences important for assembly of keratins into 10-nm filaments, we used a combined approach of (a) transfection of mutant keratin cDNAs into epithelial cells in vivo, and (b) in vitro assembly of mutant and wild-type keratins. Keratin K14 mutants missing the nonhelical carboxy- and amino-terminal domains not only integrated without perturbation into endogenous keratin filament networks in vivo, but they also formed 10-nm filaments with K5 in vitro. Surprisingly, keratin mutants missing the highly conserved L L E G E sequence, common to all intermediate filament proteins and found at the carboxy end of the alpha-helical rod domain, also assembled into filaments with only a somewhat reduced efficiency. Even a carboxy K14 mutant missing approximately 10% of the rod assembled into filaments, although in this case filaments aggregated significantly. Despite the ability of these mutants to form filaments in vitro, they often perturbed keratin filament organization in vivo. In contrast, small truncations in the amino-terminal end of the rod domain more severely disrupted the filament assembly process in vitro as well as in vivo, and in particular restricted elongation. For both carboxy and amino rod deletions, the more extensive the deletion, the more severe the phenotype. Surprisingly, while elongation could be almost quantitatively blocked with large mutations, tetramer formation and higher ordered lateral interactions still occurred. Collectively, our in vitro data (a) provide a molecular basis for the dominance of our mutants in vivo, (b) offer new insights as to why different mutants may generate different phenotypes in vivo, and (c) delineate the limit sequences necessary for K14 to both incorporate properly into a preexisting keratin filament network in vivo and assemble efficiently into 10-nm keratin filaments in vitro.