This paper uses laser-induced breakdown spectroscopy and support vector machine to analyze the content of Mn in soil. Forty-four soil samples were collected in Huaibei, Anhui. The samples were divided into training set (34 samples) and test set (10 samples) using Kennard-Stone (K-S) method. Multiple linear regression (MIR), grid search method (GSM), genetic algorithm (GA), particle swarm optimization (PSO), and least squares method (LS) were used to establish quantitative analysis models. The results show that the correlation coefficients Rtra2 of the training set of the MIR, GSM, and PSO models are only 0.861, 0.866, and 0.862, respectively. The correlation coefficients Rt2 of the test set of corresponding models are lower than 0.9, the relative error is greater than 8.6%, and the error is larger. The Rtra2 of the GA model is greater than 0.93, and Rt2 is less than 0.9. The training time of the GA model is long, so the training time must be reduced, and the correlation of the test set must be improved. The LS model works well with Rtra2 0.998 and Rt2 0.967, and the relative error is small. The training time is greatly shortened year-on-year, correlation is good, and generalization ability is strong. The LS model is more suitable for the rapid detection of the Mn element in soil.