Abstract. This paper aimed to establish the model of detecting straw composition rapidly based on OSC (Orthogonal Signal Correction) straw pretreatment. The study select soybean straw as research subjects, building predictive models for its main ingredient, namely, cellulose. Compared to the traditional denoising method respectively, calibration set model processed by the second derivative + smoothing and OSC has significantly higher determination coefficients. Applying OSC-PLS compared to the second derivative-smoothing denoising resulted in removal of non-correlated variation in spectra and improved interpretative ability of variation. Meanwhile, analysis and convergence velocity has improved significantly.