Environmental degeneration has seriously restricted the economic and social development of countries around the world. To tackle the problem, the projects of ecological restoration and reconstruction have been or are being carried out in many places. Under this background, many scholars try to assess the effects of ecological restoration through statistical method, comprehensive evaluation method, fuzzy evaluation method and grey evaluation method. However, it is difficult to discern the non-liner correlation between each assessment indicator and the degree of ecosystem restoration, as well as to decide the contribution ratio of each indicator. The methods mentioned above were complicated in assessing the contribution ratio of indicators; whereas, the back propagation neural network can solve the problems about non-linear model and contribution ratio of indicators effectively through adjusting the weight of each indicator automatically in the training process of this model. The research focuses on the small watershed of Zhuxi in Changting County, Fujian Province. The data was acquired from field investigation, lab analysis and remote sensing images which the features are extracted from. The ecosystem restoration model which can quantitatively evaluate the degree of the ecosystem restoration is built using back propagation neural network (BP-NN) by Matlab7.0 software. Firstly, four aspects covering nine indicators are chosen to assess the restored ecosystem, including soil physicochemical properties (soil organic matter, soil total N, soil total P, soil total K, soil bulk density, pH), indices of species diversity(Shannon-Wiener), thermal environment (surface temperature) and vegetation structure (vegetation coverage). The nine indicators are the input variables and the values of ecological restoration are output of the BP-NN. Secondly, the ecosystem restoration model is built by data preprocessing, sample training and sample test using Matlab7.0 software. Lastly, the ecological restoration of Zhuxi small watershed is quantitatively evaluated by the model. The results show that the predicted values from ecosystem restoration model are in accordance with the real situation, which indicates BP-NN model is feasible in quantitative evaluation of restored ecosystem. The area of extremely low ecosystem restoration in Zhuxi small watershed occupies only 0.94% and the area of medium ecosystem restoration accounts for 95.48%, which indicates that the measures of ecological protection have achieved initial results. However, the area of high ecosystem restoration accounts for only 3.62%, suggesting more work should be done in managing and protecting environment in future. The selection of assessment indicators is another core for building the model. Based on former researches, we add thermal environment data (surface temperature) to the model in this study, which can make the assessment system more comprehensive, and achieve more ideal simulation result. In further research, more indicators including biomass, composition of litter, arbor density, and height and so on will be admitted to the assessment system for more accurate result. For direct perception of the output of the model, the dot data of model output was transformed into the surface data to create the map for the degree of ecosystem restoration.