The prediction of link quality can provide link selection service for the upper-layer protocol in wireless sensor networks (WSNs). This paper presents a link quality prediction mechanism based on time series forecasting model. It mainly includes the calculation of the current value of packet reception rate (PRR), the prediction of the future value, and the routing measure of the link quality. Experimental results indicate that the mechanism can use fewer probe packets to get a more accurate estimation value and decrease extra energy consumption compared with counting-based PRR estimators.