Abstract: 【Objective】 Explore the relationship between milk dielectric characteristics and moisture content,In order to scientifically and effectively predict the moisture content of raw milk, realize rapid detection of moisture content. 【Methods】 The dielectric constant (ε') and dielectric loss factor (ε″) of milk samples between 2 and 20 GHz were measured using vector network analyzer and coaxial probe. PLSR, SVR and PSO-LSVR were used to use ε'or ε″ as variables to predict the milk from 86.9% to 92.9% and select the best, so as to achieve the accurate prediction of milk moisture content. 【Results】 The results showed that as the frequency increased, ε' gradually decreased, while ε'' gradually increased. Among the six models, ε″ based on the PSO-LSSVR method had the best water content prediction performance, with R² and RMSE of 0.996,3 and 0.001,3, respectively. 【Conclusion】 The results can provide methodological guidance for predicting other nutritional indicators of milk and a reference for studying the quality of other natural emulsions.