The cognitive radio technology has been adopted as a promised solution to deal with overcrowded problem in ISM band. In wireless sensor networks, the cognitive radio technology enables sensor nodes to occupy licensed band in opportunistic manner and introduces advantages of improving spectrum utilization and increasing system throughput. By considering the opportunistic spectrum access nature, in this paper, we aim to provide a stable path selection scheme for cognitive radio sensor networks. The proposed routing protocol utilizes priori delivery information and derives a novel routing metric for the optimal path selection. The routing metric is derived from naïve Bayes inference and used to indicate the path stability by means of delivery probability. We show through extensive simulations that the proposed routing protocol is efficient to find the most probable delivery path and increase the system throughput.