Spectral quantitative analysis for thr ee-dimensional first-order derivative fluorescence based on support tensor mac hine (STM) is proposed and applied to detect aflatoxin in foods in this paper.The conventional Savitzky-Golay polynomial smoothing and d ifferentiation methods,only used for vector-based spectral analysis,are extended to three-dim ensional fluorescence spectrometry represented by 2-order tensors.Because the three-di mensional first-order derivative spectrometry is represented as a 3-order tenor,support tensor machines (STMs) are used to build calibration model in order to improve the model prediction accurac y.In the experiments for detecting aflatoxin in liquor,the correlation coefficient (CC) and root mean square error of prediction (RMSEP) are 0.9523and 14.8475,respectively,and compared wit h partial least squares (PLS) and support vector machine (SVM) method,the CC is increased by 2.34% and 2.40% while RMSEP is reduced by 4.36% and 8.92%,respectively.And in the experiments for detecting aflatoxin in milk,CC and RMSEP are 0.9965and 5.4489,and the CC is increased by 2.34% and 2.40% while RMSEP is reduced by 4.36% and 8.92%,respectively with comparison to PLS and SVM.These experimental r esults show that the performance of calibration model used by STM is superior to that by SVM and PLS.