Volcanic tremor at the Hekla volcano is directly related to eruptive activity. It starts simultaneously with the eruptions and dies down at the end of them. No tremor at Hekla has been observed during noneruptive times. The 1991 Hekla eruption began on 17 January, after a short warning time. Local seismograph stations recorded small premonitory earthquakes from 16:30 GMT on. At 17:02 GMT, low-frequency volcanic tremor became visible on the seismograph records, marking the onset of the eruption. The initial plinian phase of the eruption was short-lived. During the first day several fissures were active but, by the second day, the activity was already limited to a segment of one principal fissure. The eruption lasted almost 53 days. At the end of it, during the early hours of 11 March, volcanic tremor disappeared under the detection threshold and was followed by a swarm of small earthquakes. At the start of the eruption, the tremor amplitude rose rapidly and reached a maximum in only 10 min. The tremor was most vigorous during the first hour and started to decline sharply during the next hour, and later on more gently. During the eruption as a whole, the tremor had a continuous declining trend, with occasional increases lasting up to about 2 days. Spectral analysis of the tremor during the first 7 h of the eruption shows that it settled quickly, within a couple of minutes, to its characteristic frequency band, 0.5-1.5 Hz. The spectrum had typically one dominant peak at 0.7-0.9 Hz, and a few subdominant peaks. Hekla tremor likely has a shallow source. Particle motion plots suggest that it contains a significant component of surface waves. The tremor started first when the connection of the magma conduit with the atmosphere was reached, suggesting that degassing may contribute to its generation.