Land-use and land-over change (LUCC) is one of the core elements of global environmental change. Large- scale and long- scale LUCC have profound effects on atmospheric composition, climate change, nutrient cycling, ecosystem, and more. The effect of human activities on the Earth has increased, especially in the past 300 years, and the resulting changes in the global environment are also profound. The reconstruction of arable land pattern over historical periods, an important part of LUCC, has been a worldwide concern in academic circles. Most of previous studies have used the total based spatial allocation approach. Taking into account the continuous distribution of arable land and spatial constraints, this paper proposes a constraint-based cellular automata model to reconstruct the historical arable land pattern. The model establishment, parameter calibration, and result validation are described in detail in this paper. We selected five constraints including soil pH value, content of soil organic matter, intensity of soil erosion, and distance to the nearest human settlements as well as distance to the nearest river, and their relationships with the arable land distribution in 1980, as the transition rule of CA, were quantitatively estimated using logistic regression. The model was applied to Jiangsu Province in China, and was compared with the conventional spatial allocation method. The results showed that the methodology developed in this study can more objectively reflect the evolution of the pattern of arable land over historical periods, in terms of similarity with contemporary pattern, than the spatial allocation methods and can provide an effective basis for the historical study of arable land. Compared to the conventional spatial allocation approach for spatial pattern reconstruction of historical arable land, this study has the following findings: (1) Borrowing ideas from urban growth simulation, constrained CA has been initially applied for reconstructing historical arable land to consider contiguous development of arable land. (2) Contemporary arable land pattern and several spatial factors were used to identify the objective transition rule of historical arable land incorporating with logistic regression, avoiding the subjectivity in some existing studies. (3) The constructed pattern can be dynamically visualized at intervals of ten years. (4) Compared with existing research, our reconstruction has high resolution (1 km grid) and is a form of land-use types (non-proportional). Reconstruction result in other coarser scale could be aggregated based on the 1-km pattern. (5) According to the characteristics of available data in the history of China, we proposed qualitative and quantitative methods to validate reconstruction results.