Between 1999 and 2004 we deployed a series of thermal infrared experiments at Stromboli volcano (Italy) to thermally define the dynamics of gas puffing. These experiments involved targeting degassing vents with thermal infrared thermometers and thermal imagers from a distance of ~300 m. The data showed that puffs involved a short-lived gas thrust phase (velocities up to ~20 m/s) followed by a dominant phase of buoyant rise with constant ascent velocities (~7 m/s), until cooling caused the puff to stall and disperse. The puff could best be described as a starting plume that rapidly developed into a thermal. Ascent velocities, puff volumes and frequencies were extremely stable over the 6-year-long study period, typically being 10 1 5 m/s, 50--190 m3 per burst and 0.5--1 Hz, respectively. These volumes convert to puff masses of 10--40 kg and give a total gas flux of 7--26 kg/s. Although stable at the system as a whole, at a single vent, gas puffing showed short-term variation and instability. Gas puffing would wax and wane at individual vents, shifting to other linked vents so that for the system has a whole, puffing activity was maintained. Variations in vent temperature can thus be tied to variations in gas flux, where termination of gas puff activity (a reduction in the flux) at a particular vent results in decreased vent temperature. We suggest that gas puffing is related to a rapid and repeated release (bursting) of gas slugs at the head of the magma column, at a rate of one every 1--2 s. Released gas packets then ascend the empty conduit section to be emitted as the puffing plume. This style of degassing activity persists between Strombolian explosions and operates alongside passive degassing; but has a very different source mechanism. It appears to represent a highly stable and repetitive, but extremely mild, stream of explosive bursts. This style of nonpassive degassing may account for up to 45% of Stromboli's total gas flux.