Wavelength model optimisation is a fundamental core technology of near-infrared (NIR) spectroscopic analysis. An integrated optimisation method for wavelength selection of serum glucose analysis was proposed in the present study. A global parameter optimisation platform of Norris derivative filter (NDF) was established to achieve the most suitable spectral preprocessing. The equidistant combination-partial least squares (EC-PLS) combined with wavelength step-by-step phase-out-PLS (WSP-PLS) was proposed to achieve large-scale screening of discrete wavelengths model for serum glucose. As a result, the combination of eight wavelengths (1583, 1695, 1733, 1816, 1860, 2125, 2340, 2376 nm) was selected and a good prediction effect was obtained. It can provide a valuable reference for non-invasive blood glucose detection and designing dedicated spectrometers. The proposed WSP-PLS can optimise any wavelength model obtained by certain optimisation strategies, improve the prediction performance and reduce the wavelength model complexity. We believe it will have a wider application.