Molybdenum, nickel, and chromium play important roles in steel property. Laser-induced breakdown spectroscopy (LIBS) assisted with laser-induced fluorescence (LIF) is a promising technique with high sensitivity to elemental analyses. However, the spectra suffered from strong and unstable background from laser scattering when determining these three elements in steel matrix, which would deteriorate the accuracy. In this work, a self-adaptive method based on discrete wavelet transform (DWT) was introduced to solve this problem. No manual or subjective intervention is needed even if changing spectral ranges and elemental species. The LIBS-LIF spectral data were decomposed by Daubechies wavelet with the wavelet function db7 and the decomposition level 7. Then the spectra were reconstituted with background removal. In quantitative analyses, R squares in calibration curves of chromium, nickel, and molybdenum were greatly increased from 0.976, 0.965, and 0.981 to 0.995, 0.993, and 0.997, respectively; and the root-mean-square errors of cross-validation (RMSECVs) were significantly decreased from 0.0153, 0.0290, and 0.0152 wt.% to 0.00649, 0.00832, and 0.00793 wt.%, respectively. The results demonstrated both calibration model accuracy and analytical accuracy were greatly improved. This work provides an effective and convenient approach for modifying LIBS-LIF analyses in determination of molybdenum, nickel, and chromium in steels.