Research is conducted to accurate and efficient algorithms for extracting ring-down time (τ) in cavity ring-down spectroscopy (CRDS) which is used to measure NO3 radical in the atmosphere. Fast and accurate extraction of ring- down time guarantees more precise and higher speed of measurement. In this research, five kinds of commonly used algorithms are selected to extract ring-down time which respectively are fast Fourier transform (FFT) algorithm, discrete Fourier transform (DFT) algorithm, linear regression of the sum (LRS) algorithm, Levenberg-Marquardt (LM) algorithm and least squares (LS) algorithm. Simulated ring-down signals with various amplitude levels of white noises are fitted by using five kinds of the above-mentioned algorithms, and comparison and analysis is conducted to the fitting results of five kinds of algorithms from four respects: the vulnerability to noises, the accuracy and precision of the fitting, the speed of the fitting and preferable fitting ring-down signal waveform length. The research results show that Levenberg-Marquardt algorithm and linear regression of the sum algorithm are able to provide more precise results and prove to have higher noises immunity, and by comparison, the fitting speed of Levenberg-Marquardt algorithm turns out to be slower. In addition, by analysis of simulated ring-down signals, five to ten times of ring-down time is selected to be the best fitting waveform length because in this case, standard deviation of fitting results of five kinds of algorithms proves to be the minimum. External modulation diode laser and cavity which consists of two high reflectivity mirrors are used to construct a cavity ring-down spectroscopy detection system. According to our experimental conditions, in which the noise level is 0.2%, linear regression of the sum algorithm and Levenberg-Marquardt algorithm are selected to process experimental data. The experimental results show that the accuracy and precision of linear regression of the sum algorithm is considerably close to those of Levenberg-Marquardt algorithm, and on the other hand, the fitting speed of linear regression of the sum algorithm is faster than that of Levenberg-Marquardt algorithm about five times. The experimental results are consistent with the simulation analysis, and it indicates that linear regression of the sum algorithm is the desirable fitting method, as far as our experimental conditions are concerned.