We report ab initio calculations of the structures, binding energies and spin multiplicities of the clusters Fe 2 , C 2 , FeC n (nϭ1 -4) and Fe 2 C n (nϭ1 -3) using a density-functional method that employs linear combinations of atomic orbitals as basis sets, nonlocal norm-conserving pseudopotentials, and the generalized gradient approximation for exchange and correlation. Our results for the pure dimers and the monometallic carbide clusters are in good general agreement with those obtained in previous theoretical studies and with available experimental data. All the dimetallic carbide clusters are predicted to have cyclic planar geometries that are stabilized ͑except, of course, in Fe 2 C) by transannular bonds. In particular, the pentagonal geometry of Fe 2 C 3 , with transannular Fe-Fe and Fe-C bonds and an FeC 2 unit that is almost identical to free FeC 2 , parallels that of Ti 2 C 3 . However, this Fe 2 C 3 structure is almost isoenergetic with another in which the C atoms aggregate to form a quasilinear C 3 substructure, as in Co 2 C 3 . This is consistent with the position of Fe in the 3d metal series, intermediate between met-car formers ͑Ti, V, Cr͒ and nonformers ͑Co, Ni͒, and with the fact that mass spectra show Fe 8 C 12 not to be significantly more stable than Fe m C n clusters of several other stoichiometries.