Abstract. We exploit self-consistent, semi-empirical molecular orbital calculations (CNDO) for large silicon clusters to characterise self-interstitials. Hexagonal (I+) and split (100) forms (I-and probably Io) are favoured among the several forms investigated. Possible extended high-temperature forms are not discussed. Our results imply Bourgoin-Corbett athermal diffusion in p-Si and low-activation energy classical motion in n-Si; local excitation enhanced motion is possible, though not verified, but local heating is unlikely. Results agree well with experiment, both for Si and in understanding the different behaviour of silicon and diamond.The energetic advantage of the split form predicted is also supported by the observed split impurity interstitials and unidentified defects, related to the self-interstitial, observed in structures similar to the split (100) form.The self-interstitial in silicon is a tantalising defect. It is one of the basic intrinsic defects, and one whose involvement is important in many solid-state processes from below 1 K to near the melting point of Si (for reviews see Watkins 1964Watkins ,1967Watkins ,1975 and Seeger et a1 1979). Yet there are no confirmed, direct observations of the self-interstitial. Several centres seen in spin resonance, internal friction or channelling might be isolated interstitials, but the consensus at present is that these are probably merely related centres. Evidence for the properties of this elusive defect relies on indirect, sometimes controversial, analyses and on the rates, natures and ranges of occurrence of solid state reactions. Our present calculations allow us to rationalise much of the lower-temperature data in a way which is consistent with corresponding but distinct results (both theoretical and experimental) for diamond.The experimental situation is one of considerable controversy. We may summarise the results concisely as follows. At the lowest temperatures (0.5-20 K), in p-Si only, replacement reactions occur which displace substitutional impurities X, into interstitial sites, the silicon interstitial moving to the vacated lattice site, i.e. Sii + X, + Xi. The long-range motion of the interstitial appears to be athermal, driven by the ionisation produced at the same time as the defect. Replacement processes are observed for X = B, Al; at higher temperatures, similar processes seem to occur for C and Ga. However, Ge is not displaced to an interstitial site. At intermediate temperatures (50-900 K) there are signs of interstitial motion in n-Si, though no evidence for athermal motion. Various studies give low motion energies in the range 0.4 eV (Watkins 1967) to 0.08 eV (Brown and Fathy 1981). At higher temperatures still (above 900K) evidence is from the