We study the problem of link quality (LQ) estimation in wireless sensor networks. Based on an extensive measurement campaign we have investigated in detail the behaviour of usual LQ estimators, such as the Packet Reception Rate (PRR), the received signal strength indicator (RSSI) and the LQI index of the CC2420 radio. All of these turn out to have shortcomings, calling for the design of a new LQ metric. We propose such a metric combining PRR and RSSI in a normalized manner. We show that for a wide variety of link conditions our metric provides stable and accurate link quality estimates. This work has a large number of potential application areas, including more accurate link-aware routing, topology control and different cross-layer optimization schemes.