This paper describes a series of laboratory experiments in which buoyant mixtures of methanol and ethylene glycol (MEG) are injected as a downward propagating jet into a tank of fresh water. As the MEG mixes with water, it becomes denser than the water. If the MEG mixes with sufficient water before its initial momentum is exhausted, the jet fluid may become dense and continue downward into the tank as a convecting plume. If the jet does not have sufficient initial momentum, then a collapsing fountain develops, and the material in the jet rises back toward the top of the tank and spreads laterally as a gravity current. Subsequent mixing can cause some of the material in the gravity current to become dense, separate from the current, and sink into the tank. Although the direction of gravity is reversed, these experiments simulate many of the important dynamical features of eruption columns which can develop during explosive volcanic eruptions. In the volcanic situation, a hot, dense, and dusty mixture of gas, ash, and clasts is erupted from a vent at high speed. If sufficient ambient air is entrained into the jet, then the mixture may become buoyant through heating and expansion of the air; it therefore continues rising high into the atmosphere. Otherwise, the material collapses back to the ground. These experiments allow one to investigate systematically the different styles of behavior of the erupted material as the eruption conditions change. Four different styles of behavior have been identified, with transitions from one style to the next as the initial momentum flux of the jet is decreased: (1) Plinian style convecting columns; (2) the periodic release of discrete convecting clouds originating close to the vent, from a collapsed fountain; (3) coignimbrite eruption columns centered some distance from the vent which are generated when a fraction of a pyroclastic flow becomes buoyant; and (4) pyroclastic flows in which the majority of the material remains relatively dense and therefore spreads laterally from the vent. Using a simple theoretical model of the laboratory experiments, an analytical expression describing the conditions necessary for collapse of the analogue laboratory columns has been derived. This is successfully compared with the laboratory experiments. The simple analysis predicts that column collapse may be induced by (1) increasing the density of the erupted material; (2) decreasing the maximum buoyancy the mixture can attain on mixing with ambient; (3) increasing the erupted mass flux for a given momentum flux; or (4) decreasing the initial momentum flux for a given mass flux, as may occur if the vent is eroded. These results are consistent with earlier studies (Wilson et al., 1980; Wilson and Walker, 1987; Bursik and Woods, 1991). Although the analogue experimental system is somewhat simplified, the observations of the periodic release of convecting thermals just after column collapse suggest a mechanism for the complex grading which is commonly found just below the collapse horizons in fall deposits, for example the Fogo A deposit (Walker and Croasdale, 1970) and the Vesuvius A.D. 79 deposit (Carey and Sigurdsson, 1987). In conjunction with partial column collapse, this periodicity also suggests a means by which flow deposits may become interspersed with fall deposits; this feature has been observed in a number of cases including the Taupo deposit (Wilson and Walker, 1985).