As an important surface drinking water source and water resource strategic reserve base in Beijing, Miyun Reservoir has attracted wide attention. Most of the relevant researches are based on watershed scale, and there is a lack of multi-scale ecological risk assessment for water resource protection areas. The multi-scale ecological risk assessment is conducive to guiding the high-quality and sustainable development of Miyun Reservoir watershed. In this study, the land use data in 1990, 2000, 2010 and 2018 were used, based on the landscape index, using geostatistical method, we analyzed the watershed land use type change, we constructed an ecological risk assessment model from the scale of the basin and water source protection area to reveal the temporal and spatial changes of ecological risk. The results showed that:(1) the main land use types in the basin were forestland and grassland. With the urban expansion and the implementation of "returning farmland to forest" policy, the area of cultivated land and unused land decreased, and the area of construction land increased. The landscape of the basin became more complex and scattered, and the degree of fragmentation intensified. (2) Ecological risk areas showed a distribution pattern of high in the edge and low in the middle at the basin scale, and the area of high-risk areas gradually shifted to low-risk areas. The high-risk areas were concentrated in Miyun District,Xinglong County and Chicheng County in the south of the basin and Fengning County in the north of the basin, and the ecological security has been gradually improved.(3) The scale of water source protection areas showed a pattern of high in the middle and low in the edge. The area of high risk gradually decreased and the spatial distribution was concentrated, and the ecological risk tended to weaken.(4) There was positive correlation between landscape ecological risk and Moran's I index, which was 0.322, 0.305, 0.298 and 0.317, respectively. Moran's I was the largest in 1990 with the strongest spatial correlation, and the "hot spot" area increased year by year, which is the key area for future watershed control and planning. Locally spatial autocorrelation distribution pattern is consistent with landscape ecological risk distribution.