The driving forces analysis is frequently used in conventional multivariate analysis such as multiple linear regression, canonical correlation analysis, principal component analysis and logistic model. These methods, however, assume that the input data are spatially independent regardless of the facts showing otherwise in landscape analysis. To overcome the shortfalls of the conventional methods to address the spatial autocorrelations, we proposed a new model (AutoLogistic) by incorporating the spatial autocorrelations into the conventional logistic model. AutoLogistic model was then applied at the Nverzhai watershed of Zhangjiajie to identify the driving forces for the cover changes. We included spatial variables (slope, altitude, aspect), distance to the nearest road, stream and residential areas in our new model. Model predictions were validated based on the Relative Operating Characteristics (ROC) method. We found that: (1) the vegetation cover changes and the driving factors appeared positive autocorrelation in space that decreases with distance across the watershed; (2) the AutoLogistic model showed higher accuracy than that of the logistic model, with remarkable reduction in the number of independent variables; (3) the magnitudes of influences from the natural driving forces seemed significantly different from those of the anthropogenic driving forces. Slope was the single most important variable on changes in farmlands, orchards, evergreen broad-leaved stands, and the coniferous stands, while aspect showed its importance for changes of farmlands, orchards, evergreen broad-leaved stands, and the shrub stands. Elevation appeared an unimportant factor driving the cover changes. Interestingly, the changes in deciduous broad-leaves stands seemed to be more influences by anthropogenic activities.