Rocky desertification is seriously restricting the sustainable development of agricultural production and tourism industry in karst regions. This study focus on the characteristics of rocky desertification and its main driving factors on non-tillage lands in Houzhai River Basin of southwestern China. Artificial neural networks (ANNs) were employed in this study to identify the importance of different environmental factors on rocky desertification. The results showed that the rock outcrops in non-tillage lands ranged from 0.00 to 91.12% with a mean value of 19.10% in the Houzhai River Basin, and the rates of rock outcrops among different types of vegetation were ordered as: shrub grasslands (35.30%)>arbor forestlands (30.07%)>shrublands (23.13%)>arbour-shrub mixed forestlands (27.54%). With increases in slope gradient and altitude, the cover rate of rock outcrops became higher and rocky desertification became more serious. Based on ANNs analysis, the correlation coefficients between observed and predicted values of remaining data sets ranged from 0.828 to 0.998, which indicated that the importance of altitude, slope gradient, gravel content and soil bulk density are the dominant factors affecting soil erosion and thereby leading to the occurrence of rocky desertification in the Houzhai River Basin. In addition, ANNs combine with environmental factors can be a feasible way to predict tendency of rocky desertification in a karst regions.