The reconstruction of land use/land cover change within small-scale areas during historical periods based on multi-source data is of great significance for obtaining an in-depth understanding of the mechanisms driving land use change and their effects on the environment and climate. Herein, we reconstructed the evolution of settlement patterns at the basin scale, the amount and spatial distribution of cropland at the settlement scale, as well as the cropland pattern at the basin scale. Typical small and medium-scale watersheds in the mountainous areas of Yunnan Province were used as the study area. Abided by historical documents, field investigation data, historical geographical research results, archival data, modern statistical data, and basic geographical data, we considered regional natural factors (slope and altitude) and anthropogenic factors (population, policies, agricultural technology, and distance between croplands and settlements) and built a gridded model for cropland reconstruction in historical periods in the small-scale mountainous areas. We then reconstructed gridded cropland patterns with clear temporal and spatial attributes for the past 300 years. The results showed that: (1) over the past 300 years, the area of cropland in the Zhangjiu River Basin has increased 6.3 times, and the reclamation rate increased from 2.1% in 1700 to 15.6% in 1978. (2) The cropland area differed considerably in various terrains. The cropland area in mountainous and semi-mountainous areas was the largest and increased rapidly, whereas the cropland area in the flatlands and the middle and lower reaches of the valley area increased slowly, which was caused by a combination of factors, such as the natural environment, population, policies, and agricultural technology. (3) The results were verified using the total cropland area and per capita cropland area. The gridded reconstruction model devised in this study can be used as a reference for simulating historical cropland in small-scale areas with distinct spatial and temporal attributes using gridded datasets.