One of the first problems that must be solved if device-quality GaAs is to be grown on Si substrates is that of suppressing antiphase disorder. Recent experimental evidence shows that such disorder can be suppressed not only on the (211) orientation, but also on (100), through a combination of slight misorientation and a high-temperature surface anneal, which lead to the pairing of all Si surface steps into a particular kind of double-height steps. A recent model by Aspnes of Ihm explains the energetic preference for this kind of step via a drastic reconstruction of the atomic configuration at the step edge involving the formation of a π-bonded chain running along the step. Another unexpected puzzle is posed by the recent observation that on a given Si (100) surface, antiphase disorder-free growth with both possible Ga–As sublattice allocations can be achieved, depending on initial nucleation conditions. A detailed nucleation model is presented that explains these observations, involving a site interchange between Si atoms and incoming As atoms under Ga-rich nucleation conditions.