Empirical analysis of Internet traffic characteristics should not be biased by the measurement methodology used to gather data. This article compares probe- (active) and router-based (passive) methods for measuring packet loss both in the laboratory and in a wide-area network. The laboratory case study demonstrates the accuracy of passive Simple Network Measurement Protocol (SNMP) measurements at low loss rates; the wide-area experiments show that active-probe loss-rate measurements don't correlate with those measured by SNMP from routers in a live network. This case study's findings also reveal that common methods for active probing for packet loss suffer from high variance and from the effects of end-host interface loss.