This study investigated a caries detection method based on the autofluorescence effect to achieve non-destructive detection of early dental caries. The spectrum characteristics of different caries were studied using spectroscopy. A fluorescence spectrometer was used to determine the excitation and fluorescence spectra of different caries samples, and the Savitzky-Golay smoothing method was used to preprocess the spectra. The excitation peak wavelengths of healthy teeth and caries were near 405 nm, and the fluorescence peak of tooth hard tissue was near 480 nm. Carious tissues had fluorescence peaks at 623 nm and 685 nm, corresponding to the peak positions of dental plaque metabolites. The fluorescence spectrum of teeth was significantly different under 405 nm light excitation, and the fluorescence image of healthy teeth was green. With increasing plaque concentration, the fluorescence image of dental caries turned red. The results show that early caries can be detected by analyzing the color difference in fluorescence images.