Abstract: To address the technical challenges of real-time online analysis of hot-dip galvanizing liquid samples in the metallurgical industry, the indicated research was proposed. The online monitoring device for laser induced breakdown spectroscopy (LIBS) was independently built. Pre-processing methods including standard normal variate, multiplicative scatter correction, second derivative (D2nd), and wavelet transform (WT) were employed to eliminate spectral scattering effects, baseline drift, and noise. Characteristic variables were selected through variable importance projection (VIP), and the VIP threshold was optimized based on five-fold cross-validation. Quantitative calibration models for magnesium and aluminum elements in hot-dip galvanizing liquid samples of Zn-Al-Mg system and Zn-Al system were established by combining partial least squares (PLS). As shown by the results, the prediction performance of the optimal spectral pre-processing-VIP-PLS model was significantly better than that of the original spectral-PLS model. The WT-VIP-PLS and D2nd-VIP-PLS models had good prediction performance for magnesium and aluminum elements in hot-dip galvanizing liquid samples of Zn-Al-Mg system, respectively. The determination coefficients (R²) of models of the magnesium and aluminum elements were 0.983 4 and 0.973 4, respectively, and the root mean square errors (RMSE) were 0.165 7 and 0.250 9, respectively. The WT-VIP-PLS model had good prediction performance for aluminum element in hot-dip galvanizing liquid samples of Zn-Al system, with R² of 0.849 7 and RMSE of 0.067 7.