Abstract: Balanced Incomplete Block-Laws of Comparative Judgement (BIB-LCJ) test was conducted to obtain the people’s visual landscape preferences with a preferential-based test for actual landscape pictures in 11 agricultural landscape partitions in Beijing. 46 undergraduates majoring in relevant environmental sciences were recruited for the test. For each picture, windows in equilateral triangular shape with 250 m side length were defined to cover different areas corresponding to the visual field. The data from the areas in the land use map of Beijing in 2006 at a scale of 1:10 000 was analyzed using landscape spatial indices. Eleven indices, including 9 landscape spatial indices, level of management and proportion of green, were used as independent variables, and multiple stepwise was used to establish the landscape visual preference models of various landscape partitions in Beijing. The analysis show that green proportion, management level, patch cohesion index, aggregation index, fractal index distribution have positive effect, and proportion of openness has negative effect on landscape visual preferences in the partitions. Patch number, contagion, largest patch index, proportion of disturbance and proportion of forest have different correlations with landscape visual preferences in different partitions.