Classifying laser-printer toner is an essential step for identifying printers and dictating forged documents. However, the existing methods require numerous training samples, which is unrealistic in document examination cases. A few-shot classification method based on infrared spectroscopy and chemometrics is proposed. The infrared spectrum of eight types of toner was collected, and optimal spectroscopic data preprocessing methods were selected according to the data’s characteristics and traverse comparison experiment. Using the processed data, a partial least square-discriminant analysis (PLS-DA) model was established. Random forest (RF) and support vector machine (SVM) were used as the comparison methods. Experimental results show that the second derivative and Savitzky-Golay smoothing are the best preprocessing methods for the collected spectrum. In all conditions, PLS-DA outperform RF and SVM. When the number of the training set is larger than 90, the accuracy of the PLS-DA model is 100%, and when the number of the training set is 60, it reduces to 95%. The discriminant model of laser-printer toner based on infrared spectroscopy and PLS-DA exhibits high accuracy and strong interpretability, requires less training samples, and can be applied in forensic science.