The explosive eruptions of July 22 and August 7, 1980, at Mount St. Helens, Wash., both included multiple eruptive pulses. The beginnings of three of the pulses-two on July 22 and one on August 7-were witnessed and photographed. Each of these three began with a fountain of gases and pyroclasts that collapsed around the vent and generated a pyroclastic density flow. Significant vertical-eruption columns developed only after the density flows were generated. This behavior is attributable to either an increase in the gas content of the eruption jet or a decrease in vent radius with time. An increase in the gas content may have occurred as the vent was cleared (by expulsion of a plug of pyroclasts) or as the eruption began to tap deeper, gas-rich magma after first expelling the upper, gas-depleted part of the magma body. An effective decrease of the vent radius with time may have occurred as the eruption originated from progressively deeper levels in the vent. All of these processes-vent clearing; tapping of deeper, gas-rich magma; and effective decrease in vent radius-probably operated to some extent. A 'relief-valve' mechanism is proposed here to account for the occurrence of multiple eruptive pulses. This mechanism requires that the conduit above the magma body be filled with a bed of pyroclasts, and that the vesiculation rate in the magma body be inadequate to sustain continuous eruption. During a repose interval, vesiculation of the magma body would cause gas to flow upward through the bed of pyroclasts. If the rate at which the magma produced gas exceeded the rate at which gas escaped to the atmosphere, the vertical pressure difference across the bed of pyroclastic debris would increase, as would the gas-flow rate. Eventually a gas-flow rate would be achieved that would suddenly diminish the ability of the bed to maintain a pressure difference between the magma body and the atmosphere. The bed of pyroclasts would then be expelled (that is, the relief valve would open) and an eruption would commence. During the eruption, gas would be lost faster than it could be replaced by vesiculation, so the gas-flow rate in the conduit would decrease. Eventually the gas-flow rate would decrease to a value that would be inadequate to expel pyroclasts, so the conduit would again become choked with pyroclasts (that is, the relief valve would close). Another period of repose would commence. The eruption/repose sequence would be repeated until gas-production rates were inadequate to reopen the valve, either because the depth of the pyroclast bed had become too great, the volatile content of the magma had become too low, or the magma had been expended. A timed sequence of photographs of a pyroclastic density flow on August 7 indicates that, in general, the velocity of the flow front was determined by the underlying topography. Observations and details of the velocity/topography relationship suggest that both pyroclastic flows and pyroclastic surges formed. The following mechanism is consistent with the data. During initial fountain collapse and when the flow passed over steep, irregular terrain, a highly inflated suspension of gases and pyroclasts formed. In this suspension, the pyroclasts underwent rapid differential settling according to size and density; a relatively low-concentration, fine-grained upper phase formed over a relatively high-concentration coarse-grained phase. The low-particle-concentration phase (the pyroclastic surge) was subject to lower internal friction than the basal high-concentration phase (the pyroclastic flow), and so accelerated away from it. The surge advanced until it had deposited so much of its solid fraction that its net density became less than that of the ambient air. At this point it rose convectively off the ground, quickly decelerated, and was overtaken by the pyroclastic flow. The behavior of the flow of August 7 suggests that a pyroclastic density flow probably expands through the ingestion of ai