In recent years, cultivated land was occupied continually, which had threatened the grain safety of city. Remote sensing is a powerful tool for protecting cultivated land by means of accurate inspecting the urban expansion and land use cover and change in time. Combined with the principle of general Monte Carlo, the relationships among cultivated land, road, urban expansion, GDP and industry economic structure were analyzed deeply using BP artificial neural networks based on TM remote sensing images, land use map and yearbook of Zhangjiagang City in several years. The results demonstrated that the cultivated land area decreased about 2.5% per year in the past ten years, however the general agriculture output had a small rebound after 2002, this attributed to the government optimized policy to land use and restructure of agriculture and industry. The expansion trends of Zhangjiagang City was the spread of multi-core along the road. Cultivated land within 2 km far from the main road and city zone was most likely changed into the urban area, especially within 500 meters. The rapid development of economic resulted in an increased transformation probability of cultivated land to urban land. However, transformation probability decreased gradually with the balance between urban land and cultivated land. So the relationship between transformation probability and GDP present an inverted “U” curve. In conclusion, BP-MC networks deal with training set of a large number of pixels, this avoids artificial neural network stepping into the local minimum point, and it is an effective method to analyze the drive factors of urban expansion quantitatively.