A spectral analysis pipeline of LAMOST (Large sky Area Multi-Object fiber Spectroscopic Telescope), which produces
archived spectral type data, is introduced. By studying observational and theoretical stellar spectra, spectral features
within medium resolution are discussed, those lines and bands with high sensitivity to stellar atmospheric parameters,
viz. effective temperature (Teff), surface gravity (logg) and metallicity ([Fe/H]), were selected. According to the
research, selected features were put into different objective algorithms to extract parameters. The application of three
algorithms to SDSS/SEGUE spectra, namely radial basis function neural network (RBFN), back propagation neural
network (BPN) and non-parameter regression (NPR), shows intrinsic statistical consistency. Based on the above research,
a stellar atmospheric parameter pipeline for LAMOST is designed.