Abstract: The traditional quantization methods lead high initial disagreement rate in the secret key generation courses, which would reduce the final length of secret key. The basic model of double-threshold quantization was proposed. The legitimate users set a high threshold and a low threshold at the same time. The sampled values are quantized to 1 when the values are greater than the high threshold and to 0 when the values are lower than the low threshold; the values between the two thresholds are not encoded. Through analysis, we can find that this method allows high reliability and does not reveal anything about the generated bits to the wiretapper. Its application in the secret key generation system was also analyzed in this paper, by taking the final secret key length generated by legitimate users as target function, the optimized quantization factors were deduced. The farther simulation results show that the legitimate users can generate more secret key bits by selecting appropriate quantization factors. The length of secret key is increased by 0.1 bits above when the signal to noise is between 15dB to 25dB.