Abstract: Abstract: With the improvement of urban economic development, leisure agriculture becomes one of the important forms of the integration development of rural and urban region. The distributions pattern of the leisure agriculture could be studied in multi-scale (from local scale to regional scale). Based on regional economic statistical data, recent researches exploring the spatial distribution pattern of leisure agriculture were difficult to accurately describe because the analyzing scale was too large. Recently, point of interest (POI) data has been widely applied to analyze features of the urban to show the spatial distribution patterns with advantages of large data volume, high position accuracy and the clear classification of the leisure agriculture format. While the researches on the distribution pattern of the leisure agriculture based on POI were rather inadequate. The key to apply kernel density estimation(KDE) for spatial distribution pattern of urban point element was exploring the suitable bandwidth. So far, most of researches on bandwidth determination of KDE were based on mathematical or visual interpretation methods. Considering the scale features of spatial distribution pattern, basing on POI data, applying the method of spatial autocorrelation to determine different bandwidths of KDE which suited to analyze the regional-scale and local-scale spatial distribution patterns, this paper has recognized and analyzed the multi-scale features of spatial distribution patterns on Beijing leisure agriculture. The spatial autocorrelation method contained 2 indices, Moran'I and Local Moran'I. With increasing of KDE bandwidth, the value of the normalized Moran'I gradually increased from 0 to 1 and the value of the normalized number of HH (a statistically significant cluster of high values) region from the result of local Moran'I decreased from 1 to 0. At this time, the result of KDE could express the regional scale distribution pattern of leisure agriculture and the detail features were covered. On the contrary, the result of KDE could express the small scale pattern of Beijing leisure agriculture and the detail features were obvious. Considering the variable characteristics of Moran'I, the number of HH and comprehensive I curves, this paper determined different bandwidths of KDE which suited to analyze the regional-scale and local-scale spatial distribution pattern of the leisure agriculture were 9 km and 3 km. Using the method of natural breaks, the results of KDE with 9 km or 3 km bandwidths were divided into three levels, high value, middle value and low value respectively. Then spatial distribution patterns with different scales were identified. Under the regional-scale, the spatial distribution pattern showed two rings and multi-centers. The interior ring away from the city center from 30 km to 50 km was one day tour zone. The exterior ring away from the city center from 50 km to 90 km was two days and up tour zone. It was found that a primary distribution center was lied in the sub-montane area in the east of Changping County, the southeast of Huairou County, the southwest of Miyun County. Two sub-centers were lied in the northeast of Miyun County and the southwest of Fangshan County. Under the local-scale, the spatial distribution pattern showed 26 centers divided into 3 classes by area size. Most centers lied in the sub-montane area in the north to the middle part of Beijing. Fewer lied in the rural-urban fringe area in the southeast and hills in the west of Beijing. The least lied in the central city area. Finally the literature analysis method was applied to certificate the multi-scale features of the spatial distribution pattern on Beijing leisure agriculture. The method proposed in this paper is effective to express the different scales of spatial distribution pattern on Beijing leisure agriculture.