Temperature is one of the most important parameters in studying of laser-induced plasma characteristics. To reduce the impact of continuous background on the calculation of temperatures using Boltzmann plots, the wavelet transform was used to decompose the spectrums, and the low-frequency signals represented the spectral baseline were deducted by using soft-threshold method. Selecting the appropriate wavelet decomposition level L and threshold coefficient α can increase the linear regression coefficient R2 of Boltzmann plots, and the calculation accuracy of plasma temperature was improved. The LIBS spectra of low alloy steel sample region from 417 to 445 nm were decomposed by using db4 wavelet, and then baseline subtraction and signal reconstruction were carried out, respectively. Twelve Fe atomic lines were chosen to establish Boltzmann plots, and the temperatures were calculated from the slope of the fitted lines in the plots. The value L and α were optimized according R2, the results showed that the 8-layer db4 wavelet decomposition can gain the high R2, while the value of α associated with the delay time td, e. g., the optimum α corresponding to maximum values of R2 is 0.3 when td≤4.0 μs, and then decrease with the increasing of td, and reduced to 0 when td≥6.0 μs. The interference due to baseline on the spectral characteristic lines gradually reduced with the increasing of td, and therefore α decreased with td increase. After the baseline was deducted, the temperature calculated by Boltzmann plot decrease of about 2 000 to 3 000 K. The temperature gradually decreased with the increasing of the td, and the temperature fluctuation is reduced after baseline subtraction, these results are consistent with the physical process of plasma expansion.