A new model calculation method,local weighted partial least squares(LWPLS),was proposed to solve the problem of nonlinear phenomena in near infrared spectroscopy.Taking Antai pill as the research object,the quantitative model for Antai pill was established by LWPLS algorithm,and the accuracies of quantitative models established by PLS and LWPLS were compared.Results indicated that,the correlation coefficient(R2) of PLS and LWPLS algorithms for ferulic acid were 0.785 5 and 0.971 9,the root mean square errors of prediction(RMSEP) were 0.126 6 and 0.043 8,and the relative prediction errors(RE) were 12.66% and 9.18%,respectively,while the R2 values of PLS and LWPLS algorithms for senkyunolide A were 0.886 4 and 0.964 9,the RMSEP values were 0.114 8 and 0.077 1,and the RE values were 14.01% and 7.81%,respectively,which showed that LWPLS algorithm was more accurate.Therefore,the LWPLS algorithm could improve the accuracy of quantitative model for Antai pill,exhibiting a wide generalization and application potential.