In order to explore a rapid and accurate detection method for honey containing uncine-kisses, mid-infrared(MIR) spectral images of 216 non-toxic and uncine-toxic honey samples from different regions are collected by MIR spectrometer. The 800~1 800 cm-1 band of the spectral images is selected as the main research band. Principal component analysis and partial least square regression(PLS-DA) algorithm are used to establish the identification model of hook-free and non-hook-free, and the prediction results of different data preprocessing methods are compared. The simulation results show that the identification model established after the first derivative pretreatment has the best identification effect, and the identification accuracy reaches 90.74%.