Dental caries has attracted widespread concern because of its universality. However, the current diagnosis of dental caries depends largely on naked eyes and dentists' experience, which causes considerable randomness and inaccuracy in the diagnosis. In this study, temporally shaped double pulse (DP) femtosecond laser-induced breakdown spectroscopy (LIBS) has been used for tooth composition detection for the first time. Under the laser fluence of 20 J/cm2, the spectra intensity for DP LIBS with a pulse delay of 200 ps was 13 times higher than that of single pulse. For further prediction, the detection model based on DP LIBS and partial least squares discrimination analysis could discriminate dentin, enamel, and caries with a prediction accuracy of above 90%. This method may provide rapid and accurate feedback to dentists for real-time diagnosis in the future.