We investigate the NLTE formation of the silicon spectrum in one-dimensional semi-empirical models of the solar atmosphere HOLMUL, MACKKL and VAL,C. We derive the NLTE abundance of silicon from equivalent widths of 65 Si I lines of different strengths in the aforementioned models. We show that the abundances fitted from these lines do not depend neither on their excitation potentials, nor on their wavelengths, nor on their equivalent widths when employing Unsöld’s approximation for determination of van der Waals broadening constant γ₆ with an enhancement factor of E = 1.5. In that case the average silicon abundance range from 7.547±0.012 with HOLMUL to 7.582±0.013 with VAL,C. Using semi-classical theory of Anstee, Barklem and O’Mara for calculation of γ₆ causes reduction in the abundances with increase of the equivalent widths. We determine the NLTE silicon abundance corrections. They turn out to be on average around –0.05 dex. We test the effects on the fitted abundance resulting from uncertainties in the photoionization, inelastic collisional electron and hydrogen cross-sections. The average abundance correction caused by these uncertainties does not exceed 0.02 dex.We estimate the abundance differences due to using shifted “solar” oscillator strength scale of Gurtovenko and Kostik instead of experimental oscillator strength scale of Becker et al. We show that both scales produce almost the same silicon abundance.