Physical modelling is usually a complicated way to estimate soil moisture content, while machine learning algorithms have the potential to retrieve information from remote sensing data. In this study, the neural network, one of the most common machine learning algorithms, was used to map soil moisture from active microwave and optical data in combination. The study area was set in the middle stream of Heihe River Basin in China, from where Landsat and Envisat ASAR data were acquired in July 2008. The neural networks were trained with ground truth data and input parameters extracted from remote sensing data including bands information, Normalized Difference Vegetation Index (NDVI), Brightness Index (BI), the dual polarizations (HH and VV) and the ratio (HH/VV). Compared to an existing output of an empirical model with purely Envisat ASAR data in the same area (with R²=0.71), this study showed a slightly better correlation between the measured and estimated soil moisture (R²=0.75). It also revealed that the model with multi-source data had a better performance than the one with only a single source data, and that the selection of input parameters and the number of hidden layers and nodes can also affect the model's accuracy. Finally, the verified model was applied to the whole study area, and it was showed that this method has operational potential for estimating soil moisture under vegetated area in the middle stream of Heihe River Basin.