Rapid variability on a time-scale much faster than the light-crossing time of the central supermassive black hole has been seen in TeV emission from the blazar PKS 2155−304. The most plausible explanation for this puzzling observation is that the radiating fluid in the relativistic jet is divided into a large number of subregions which move in random directions with relative Lorentz factors . The random motions introduce new relativistic effects, over and above those due to the overall mean bulk Lorentz factor Γb of the jet. We consider two versions of this ‘jets in a jet’ model. In the first, the ‘subjets’ model, stationary regions in the mean jet frame emit relativistic subjets that produce the observed radiation. The variability time-scale is determined by the size of the subregions in the mean jet frame. This model, which is loosely based on magnetic reconnection, has great difficulty explaining the observations in PKS 2155−304. In the alternate ‘turbulence’ model, various subregions move relativistically in random directions and the variability time-scale is determined by the size of these regions in their own comoving frames. This model fits the data much more comfortably. Details such as what generates the turbulent motion, how particles are heated, and what the radiation process is, remain to be worked out. We consider collisions between TeV photons and soft photons and find that, in both the subjets and turbulence models, the mean bulk Lorentz factor Γb of the jet needs to be >25 to avoid the pair catastrophe.