Evaluation indexes of surrounding rock stability are usually in forms with normal distributions in finite intervals, and their classification boundaries are of fuzzy characteristics. So it is hard to interpret these characteristics using a normal cloud model. A novel cloud model coupled with connection number is described here to take into account of the fuzziness and conversion situation of classification boundary, and interval nature of evaluation indicators for surrounding rock. In the model, the connection cloud model in finite intervals is first discussed. Then combined with index weights, a concept of contact probability is introduced to specify the comprehensive certainty degrees of the measured evaluation indicator to various classification grades. The results from case study and comparison with other methods show that the model proposed is feasible and effective; and it can overcome the shortcomings of inability to simulate quantitatively certainty and uncertainty relationships among indicators of interval distribution form from a unit way, so as to extend serviceability for the cloud model.