Since the first crystal structure of an immunoglobulin revealed a modular architecture, the characteristic β-sheet fold of the immunoglobulin domain has been found in many other proteins of diverse biological function. Here, a systematic comparison of 23 Ig domain structures with less than 25% pairwise residue identity was performed using automatic structural alignment and analysis of β-sheet and loop topology. Sequence consensus patterns were identified for nine distinct families with at most marginal similarity to each other. The analysis reveals a common structural core of only four β-strands (b, c, e and f), embedded in an antiparallel curled β-sheet sandwich with a total of three to five additional strands (a, c^I, c^II, d, g) and a characteristic intersheet angle. The variation in the position of the edge strands (a, c^I, c^II, d and g ) relative to the common core defines four different topological subtypes that correlate with the length of the intervening sequence between strands c and e, the most variable region in sequence. The switch of strand c^I from one sheet to the other in seven-stranded domains appears to result from short c-e segments, rather than being a major structural discriminator. The high degree of structural flexibility outside the common core and the extreme variability of side-chain packing inside the core do not support a protein folding pathway common to all members of the structural class. Mutation rates of immunoglobulin-like domains in different proteins vary considerably. Disulfide bridges, thought to contribute to structural stability, are not necessarily invariant in number and location within a subclass.