In this paper, we apply the Gilbert-Elliot model to analyze measurements on a large-scale wireless sensor testbed with nodes equipped with CC1101 and CC2420 radio chips. The model considers a packet loss process as a sequence of transitions between Good and Bad states. We analyze the Packet Reception Ratio (PRR) based on the probabilities derived from the Gilbert-Elliot model. We show that the probabilities and other parameters (run and loss lengths) can very well discriminate between the main categories of link quality such as good and weak links. Identification of link quality is crucial for routing protocols such as RPL and LOADng.