Power-saving is a critical issue for Wirelesssensor networks (WSNs). A Learning-based power effcientrouting (LPER) algorithm is proposed. In the LPER, afitness function, which balances network lifetime, energyconsumption, and packet delay, is constructed and used inan ant colony system to establish the optimal route. Inaddition, reinforcement learning is applied in predictingthe energy consumption of neighboring nodes. The LPERis able to optimize network lifetime of WSNs, while keeping energy consumption and packet delay in a relative lowlevel. Numeric experiments show the LPER outperformsthe Minimal spanning tree (MST) and the Least energytree (LET) based routing algorithms in terms of networklifetime and packet delay.