The mixture of gasoline, kerosene and diesel is taken as the research object, in which diesel is considered as interferent, to analyze the content of gasoline and kerosene in the mixture. A new method is proposed to determinate the ingredient in the oil mixture and corresponding content of the component by analyzing the three-dimensional fluorescence spectral data. The method combines unfolded partial least-squares (U-PLS) with residual quadrilinearization (RQL) to process the first-order derivative three-dimensional fluorescence spectra of the oil mixture, which are extended to five-dimensional derivative spectra by using the Savitzky-Golay polynomial fitting differential to calculate the partial derivatives of the x axis and y axis of the spectral data. The calibration model of the fourth-order data is established by U-PLS/RQL to analyze the samples to be predicted, and the spectral data are decomposed and identified reasonably. The relative error of prediction is reduced to less than 5.0%, and the prediction accuracy is improved compared with that of the third-order multivariate calibration method.