Abstract: To solve the problem of input variables selection in support vector regression（SVR） model of river and lake water level, a synchronized optimization method of model structure-parameter based on envolutionary algorithm was proposed. The method can deal with the high-dimensional characteristics of the input variable search space of the complex river-lake water system, and reduce the model error resulting from the model structure and parameter uncertainties. The proposed method was applied to water level modeling at Chenglingji in Dongting Lake and Chenerkou in Jingjiang River. The results show that the proposed method can accurately reflect the effects of different influencing factors on water level prediction. The main external variables are discharged from the Yangtze River and Xiang River for Chenglingji, while thewater level prediction of Chenerkou is Zhicheng station and Majiadian water level; the method can fully exploit the potential of SVR, and the water level models at both stations achieve ideal accuracy（R₂＞0.998）; the n-fold cross-validation method used in the proposed can effectively avoids the model over-fitting problem. In summary, the proposed SVR model structure-parameter synchronous optimization method is suitable for water level modeling of rivers and lakes, especially complex river-lake water system.