Restricted to the limited battery power of nodes, energy conservation becomes a critical design issue in underwater acoustic sensor networks (UW-ASNs). Transmission with excess power not only reduces the lifetime of sensor nodes, but also introduces immoderate interference. It is ideal to send packets with moderate transmission power. In this paper, we propose a multilevel power adjustment (MLPA) mechanism for UW-ASNs to prolong the individual node lifetime and the overall network lifetime. The energy conservation is achieved by reducing the average transmission power. The analytical model is built for the MLPA mechanism enabled with k distinct power levels (k-LPA). For β=3, the closed-form expression of optimal power setting is derived and the average transmission power can be minimized as (k+1)/2k of original fixed power. For arbitrary β, a recursive formula expression is established to acquire the optimal power configuration and the minimum average transmission power as well. Furthermore, to reduce the computing complexity and the effort of parameter measurement, two approximate power configuration methods are proposed. The analytical results show that both the proposed approximate methods are near-optimal solutions for various β.