A consistent theory is presented of a mechanism for unusual photostructural changes observed in glassy semiconductors. The theory takes into account that in these materials negative-U centers are the basic charge carriers of which the electron ground state is formed by strong self-trapping of a singlet electron pair in a soft atomic mode in a glass. As a gap-light generated excited state of a negative-U center is related to a large atomic displacement in the soft mode, it is shown that metastable "defects" can indeed be created in the original structure. These are identified as the photostructural changes. The theory provides consistent answers to open basic questions. First calculations of the transition probabilities related to this phenomenon are presented.Introduction As is observed in many experiments [1, 2], light of frequency comparable to the frequency of the optical gap, w % E opt = h, can produce long-lived "photostructural changes" (PSC) in glassy semiconductors (GS), particularly in chalcogenide glasses, with the mobility-gap width E g % ð1 À 3Þ eV and E opt ' E g . Related substantial changes are produced in a variety of macroscopic physical and chemical properties of these glasses. A surprising effect was the "photodarkening" effect, gap-light induced decrease DE opt of the original optical gap width E opt by up to around 10%. Interesting anisotropic effects (e.g., dichroism, birefringence) associated with the PSC have also been revealed recently [1]. The nature of the PSC was one of the challenging problems in the physics of GS since they were discovered about three decades ago. A number of models, involving different assumptions concerning microscopic mechanisms, have been proposed to account in general for photo-induced metastable effects in semiconductors and insulators [3], in which the metastability was associated with generation of electronic excitations. However, as noted in some papers (e.g., [1], p. 481), such models, based on the assumption that adiabatic potentials of the local atomic configuration in an important atomic motion mode were different in its ground and light excited electron states, were merely plausible suggestions which accorded with known features of the effect. In fact, even for crystalline materials such models have been put on a proper theoretical footing only for defect generation in alkali halides and crystalline SiO 2 . In other cases, the important atomic mode, as well as the parameters of the local atomic configuration, actually were not well defined. For the PSC [1, 2, 4], such models postulated that the adiabatic potential in its ground and excited electron states consisted of a lower branch as a double-well potential (DWP) and an upper branch as a single-well potential, separated by a large split energy D with D= h ) w v , the vibration frequency. The process of creation and destruction of metastable defects was supposed to be due to competing