Contention-based Geographic Forwarding (CGF) has emerged recently as an energy-efficient communication proto- col to improve the wireless sensor networks lifetime. By exploiting the wireless broadcast advantage and node spatial diversity, it mitigates the effect of varying channel condition and node unavailability. The main contributions are based on classical relay selection scheme such as most forward within radius. However, this relay selection scheme performs poorly under real radio environments because it tends to forward packets on unreliable link, which leads to high expected number of retransmissions in each hop and an important number of duplicated packets. In this paper, we investigate the performance of several relay selection schemes. First, extensive simulations are proposed to evaluate their performance locally in terms of packet delivery ratio and duplicated packets. Then, we extend the work in multi- hop wireless networks and evaluate their performance in term of energy. To highlight their gain, we propose a smart solution obtained by linear programming. Based on the intelligent relay selection scheme, we develop a new solution, Furthest Forward within Reliable neighbors (FFRe), which tries to save more energy than the most efficient relay selection schemes.