Tunable Semiconductor Laser Spectroscopy (TDLAS) is a rapidly developing spectroscopic detection technology, which is widely used to detect flammable and explosive hazardous gases in the industrial environment. These hazardous gases are often mixed by various gases. The direct absorption spectrum of gas changes under different background gases which causes errors in the calculation. This paper proposes a new algorithm of mixed gases under different background gases. The algorithms of mixed gas concentration under different background gases are discussed, the reasons for Lorentz absorption spectrum changing under different background gases are studied. The reasons for errors of peak algorithm and integral algorithm in calculating mixed gas concentration are analyzed. A new method is proposed to fit the Lorentz absorption spectrum by Levenberg-Marquardt algorithm and to characterize the concentration of mixture by the quadratic fitting of area coefficient and measured concentration. In the experiment, a gas detection system based on TDLAS technology was built. The laser with a center wavelength of 1 368.59 nm was used. The length of the gas chamber was 30 cm. The water vapor was used as the gas to be tested. Dry air, nitrogen and argon were used as the background gas. The humidity generator GRZ5013 produces a relative humidity environment of 40% to 80%. Using the measurement results of the Mitchell-s8000 dew-point meter as a reference value, the quadratic fitting relationship between the three parameters (the peak, integral and area coefficients of the dry air as the background gas) and the water vapor concentration of the Mitchell dew-point meter is obtained. The relative errors of the three algorithms in the background of nitrogen and argon were compared. Experiments show that the maximum relative error of the peak, integral and area coefficient algorithms under nitrogen are -11.64%, 2.65% and 1.76%, the minimum errors are -7.79%, -0.56% and -0.54%, and the relative error mean square values are 0.88%, 0.03% and 0.01%. The maximum relative errors of the peak, integral and area coefficient algorithms under argon are -109.27%, -10.13% and 2.96%, and the minimum errors are 73.04%, -5.51% and 1.34%, and the relative error mean square values are 87.51%, 0.61% and 0.06%. The area coefficient is the best concentration algorithm in three with the smallest error and the most accurate results.