Assessing the environmental impacts of agricultural production systems requires spatially explicit information regarding cropping systems. Protecting changes in agricultural land use that are caused by changes in land management practices for analyzing performance of land activity-related policies, such as agricultural policies, also requires this type of data for model inputs. Crop sequences, which are vital and widely adopted agricultural practices, are difficult to directly detect at a regional scale. This study presnets innovative stochastic data mining that was aimed at describing the spatial distribution of crop sequences at a large regional scale. The data mining is performed by hidden Markov models and an unsupervised clustering analysis that processes sequentially observec (from 1992 to 2003) land-cover survey data on the French mainland named Teruti. The 2549 3-year crop sequences were first identified as major crop sequences across the entire territory, which included 406 (merged) agricultural districts, using hidden Marko models. The 406 (merged) agricultural districts were the grouped into 21 clusters according to the similarity of the probabilities of occurences of major 3-years crop sequences using hierarchical clustering analysis. Four cropping systems were further identified : vineyard-based cropping systems, maize monoculture and maize/wheat-based cropping systems, temporary pasture and maize-based cropping systems and wheat and barley-based cropping systems. The modeling approach that is presented in this study provides a tool to extract large-scale cropping patterns from increasingly available time series data on land-cover and land-use. With this tool, user can (a) identify the homogenous zones in terms of fixed-length crop sequences across a large territory, (b) understand the characteristics of cropping systems within a region in terms of typical crop sequences, and (c) identify the major crop sequences of a region according to the probabilities of occurences.