Sky subtraction is a key technique in data reduction of multi-fiber spectra. Knowledge of characteristics related to the instrument is necessary to determine the method adopted in sky subtraction. In this study, we describe the sky subtraction method designed for the Large sky Area Multi-Object fiber Spectroscopic Telescope (LAMOST) survey. The method has been integrated into the LAMOST 2D Pipeline v2.6 and applied to data from LAMOST DR3 and later. For LAMOST, calibration using sky emission lines is used to alleviate the position-dependent (and thus time-dependent) ∼ 4 % fiber throughput uncertainty and small wavelength instability (0.1 Å) during observation. Sky subtraction using principal component analysis (PCA) further reduces 25% of the sky line residual from OH lines in the red part of LAMOST spectra after the master sky spectrum, which is derived from a B-spline fit of 20 sky fibers in each spectrograph. Using this approach, values are adjusted by a sky emission line and subtracted from each fiber. Further analysis shows that our wavelength calibration accuracy is about 4.5 km s-1, and the averages of residuals after sky subtraction are about 3% for sky emission lines and 3% for the continuum region. The relative sky subtraction residuals vary with moonlight background brightness, and can reach as low as 1.5% for regions that have sky emission lines during a dark night. Tests on F stars with both similar sky emission line strength and similar object continuum intensity show that the sky emission line residual of LAMOST is smaller than that of the SDSS survey.