Rapid quantitative determination of the illegal addition of lambda-cyhalothrin in Bacillus thuringiensis preparations was carried out using attenuated total reflection Fourier transform infrared spectroscopy (ATR-FTIR) coupled with partial least squares (PLS) method. Three different sources of Bacillus thuringiensis preparations with different masses of 95% (w/w) lambda-cyhalothrin prodrug were added to prepare 153 mixed samples, concentrations ranging from 0.00% to 5.00%, acetonitrile was used as the extractant, and less extractant was used to increase the concentration of lambda-cyhalothrin in the extract to enhance the infrared absorption. Three pretreatment methods (smoothing, standard normal variation (SNV) and multiple scattering correction (MSC)) and six variable selection algorithms (uninformative variable elimination (UVE), interval partial least squares (iPLS), moving window partial least squares (MWPLS), competitive adaptive reweighted sampling (CARS), the bootstrapping soft shrinkage (BOSS) and interval combination optimization (ICO)), were used to examine the effects of different pretreatment methods and variable selection methods on the model results. Among them, the MSC pretreatment method combined with the BOSS algorithm obtained the optimal model results, with this method, RMSECV=1.18×10^-3, $R_{\text{cv}}^2$=9.94×10^-1, RMSEP=1.01×10^-3, and $R_{\text{pre}}^2$=9.93×10^-1. For samples with lambda-cyhalothrin concentrations ranging from 0.10% to 5.00%, the average relative error of external test samples was 4.44%, and for samples with concentrations ranging from 2.00% to 5.00%, the average relative error of external test samples was only 2.64%. This method could be applied to rapidly detect the illegal addition of lambda-cyhalothrin in Bacillus thuringiensis preparations.