Carotenoids are one of the most important components to reflect the physiological state of plant, such as environmental stress, photosynthesis as well as growing state. An effective and nondestructive detection method for the contents of carotenoids in Longjing-43 leaves was established based on Raman spectroscopy in this research. A total of 315 tea sample were used for the spectral collection and spectrophotometry determination. Quantitative models were established to predict the content of carotenoids. Firstly, in order to eliminate the interference from noise, 5 preprocessing methods were applied before calibration stage. Partial least square regression (PLS) was applied as calibration method to establish the Raman spectra quantitative model of the carotenoids in Longjing-43’s leaves. The results of PLS models based on these methods were used to evaluate the performances of these pretreatments. The best performance was achieved with wavelet transformed spectra, obtaining correlation coefficient (r) value of 0.817 and 0.786 for validation and prediction, respectively. Secondly, to further explore the Raman spectral response properties of carotenoids in tea, Successive Projections Algorithm (SPA) was applied to extract the characteristic wavenumbers for carotenoids in tea. Then, the selected 17 wavenumbers were used to build PLS model and good results are reached, obtaining Rv of 0.808 and Rp of 0.777. Based on the PLS model, this article studied the carotenoids content of tea leaves in 4 different growth periods. Results indicated that the content of carotenoids was firstly increased then decreased. The above results revealed that it was feasible to apply Raman spectroscopy for the determination of the carotenoids content in Longjing-43 leaves.