Recently we generated keratin 10 knockout mice which provided a valuable model for the dominantly inherited skin disorder epidermolytic hyperkeratosis. Here we investigated the molecular basis for their phenotype. Hetero- and homozygotes expressed a truncated keratin 10 peptide which has been identified directly by microsequencing. Epitope mapping of monoclonal antibodies to keratin 10T enabled us to study its distribution relative to keratin 6, which is highly expressed in keratin 10 knockout mice, by double-immunogold electron microscopy. This revealed that keratin 10T was restricted to complexes with keratin 1 but did not mix with keratin 6. The latter did not form extended filaments with keratins 16/17 but aggregates. Keratins 6/16 were unable to compensate for the lack of normal keratin 1/10 filaments. Remarkably keratin 6 aggregates strictly colocalized with keratohyalin granules. Residual keratin 1/10T clumps were located in the cell periphery and at desmosomes which maintained a normal architecture. Surprisingly keratin 2e, a keratin tailored to sustain mechanical stress, was completely lost in paw sole epidermis of homozygous keratin 10 knockout mice, pointing to keratin 10 as its partner. The selective pairing of keratin 10T and the loss of keratin 2e indicate that in vivo keratins are less promiscuous than in vitro. Skin fragility in keratin 10 knockout mice and in epidermolytic hyperkeratosis is probably the consequence of two complementing mechanisms namely a decrease of normal keratin 1/10 filaments and an increase in keratins 6/16 with a poor filament-forming capacity.