Abstract: The unique hydrological conditions of lakeside wetland are different from other ecosystems, and their environmental factors change frequently and have great influence on the speciation of contaminants in the sediments. The Yangzonghai lakeside wetland located in Yunnan Province was selected as the research object, and sediment samples were collected. The contents and distribution characteristics of environmental factors, total arsenic (S-TAs) and each species of arsenic of these samples were determined, and correlation analyses between species of arsenic and environmental factors were conducted. The concentrations of four species of arsenic in sediments as functions of environmental factors were predicted and compared based on both the stepwise regression and back-propagation network (BP neural network) models. The results showed that the concentrations of total arsenic in water (W-TAs) and sediment (S-TAs) were slightly higher in summer, and the contents of arsenic were near the level Ⅲ value and level Ⅴ value of environmental quality standard for water. The contents of other physical factors (pH, DO, E_h and TDS) were higher in winter, and there were obvious seasonal differences among sediment pH (S-pH), dissolved oxygen (DO) and oxidation-reduction potential (E_h). The proportion of active arsenic forms (weak acid extracted, reducible and oxidable arsenic) in the sediments of the lakeside wetland ranged from 17.70% to 62.59%, and near 80% of the sampling points showed relatively low contents of active arsenic, correspoding to low risk of ecological hazards. In the different seasons, the lakeside wetland had a significant retention effect on arsenic, and the correlation analyses showed that S-pH, DO and E_h had great influence on the distribution of arsenic speciation. Compared with the stepwise regression model, the BP neural network model was applied through the calculation of the input layer to the output layer, which enhances the ability of nonlinear and adaptive processing. The fitting degree of measured values and predicted values was up to 0.9995, while that of the stepwise regression was only 0.3749. Thus, the BP neural network model could better reflect the relationship between arsenic speciation and environmental factors and could more accurately predict their contents. The study showed that the environmental factors of the lakeside wetland had a certain influence on the occurrence forms of arsenic. Due to the complex nonlinear relationship between arsenic speciation and environmental factors, the BP neural network model was more accurate than use of mathematical statistics.