We propose a model for the expansion and hadronization of small droplets of quark-gluon plasma assumed to be produced in ultra-relativistic collisions of hadrons and/or nuclei, and we discuss its experimental consequences. The deconfinement transition is described in terms of formation of QCD strings in the expanding plasma droplet, the string tension stopping the expansion and string breaking leading to possible subdivision of the droplet into smaller ones. The bulk of plasma hadronization is predicted to occur by deflagration of the plasma droplets at transition, through their outer surface and with ejection of low pressure hadron gas. In the rapidity distributiondn/dy of hadrons, this predicts isolated maxima of width δy∼1 or high bumpy regions a few units of rapidity wide, on an event-by-event basis, the corresponding particles having larger than normal transverse momentum. It is in these rapidity intervals that one expects the “traditional” signals for plasma formation (direct dileptons, direct photons, strange particles). We review possible evidence for plasma formation in existing cosmic ray and hadron collider data. Although it appears to be inconclusive, the observed features of the rapidity distribution and the estimated energy densities in the central region are quite encouraging for a systematic search using all proposed signals.