Energy harvesting cognitive radio sensor networks (EH-CRSNs) are an emerging technology for low-cost, green monitoring of a wide range of environments. How to analyze the transmission capacity is a fundamental and challenging problem in EH-CRSNs due to the dynamics of spectrum and energy arrivals. In this paper, transmission capacity analysis of underlay relay-assisted EH-CRSNs is considered, where some nodes serve as decode-and-forward relays to assist the communication between one secondary source and one destination node. To characterize the end-to-end performance of underlay relay-assisted EH-CRSNs, we first assume that EH devices use harvest-store-use (HSU) mode and formulate the battery states with M/M/1/c model for the arbitrary integer value of transmission energy level threshold. Then, the closed-form expressions of transmission capacity are derived for the random, the nearest, and the farthest relay selection, respectively, based on stochastic geometry. In addition, the transmission capacities with the variable source-destination distance are also analyzed for the three kinds of relay selections. Finally, numerical simulations show that the transmission capacity of underlay relay-assisted EH-CRSNs can be influenced by large amounts of factors, including secondary access probability, source-destination distance, signal-to-interference ratio, and battery transmission energy level threshold, and the different relay selection schemes. The results also confirm that the random relay selection, compared with the nearest and farthest relay selection, is a more feasible and reasonable scheme for short range of underlay EH-CRSNs due to its low complexity implementation.INDEX TERMS Energy harvesting, cognitive radio sensor networks, relay selection, transmission capacity, access probability.