A method combining finite difference method (FDM) and k-means clustering algorithm which can determine the threshold of rock bridge generation is proposed. Jointed slope models with different joint coalescence coefficients (k) are constructed based on FDM. The rock bridge area was divided through k-means algorithm and the optimal number of clusters was determined by sum of squared errors (SSE) and elbow method. The influence of maximum principal stress and stress change rate as clustering indexes on the clustering results of rock bridges was compared by using Euclidean distance. The results show that using stress change rate as clustering index is more effective. When the joint coalescence coefficient is less than 0.6, there is no significant stress concentration in the middle area of adjacent joints, that is, no generation of rock bridge. In addition, the range of rock bridge is affected by the coalescence coefficient (k), the relative position of joints and the parameters of weak interlayer.