Simple arguments show that ascending thermal plumes will entrain their surroundings as the result of coupling between conduction of heat and laminar stirring driven by the plume motion. In the initial stages of ascent of a plume fed by a continuous buoyancy flux (a starting plume) the plume consists of a large buoyant head followed by a narrow vertical conduit. Laboratory experiments reported here show that the spherical head entrains ambient material as it rises, while the axial conduit carries hot source material to the stagnation point at the cap of the plume, from where it spreads laterally into thin laminae. We develop an analysis of the effects of entrainment on the structure and dynamics of starting plumes. The analysis predicts that under conditions appropriate to the earth's mantle large volumes of cooler lower mantle will be stirred into the head of a plume by the time it reaches the top of the mantle if it originates at the core-mantle boundary. The result is a major cooling and enlargement of the head. Source material ascending in the trailing conduit will undergo little contamination or cooling until the conduit is deflected from the vertical by large scale shear associated with plate motion. This plume structure explains the close association of high-temperature melts (komatiites or picrites) with more voluminous, lower temperature basalts in Archaean greenstones and modern continental flood basalt provinces: the picrites can be produced by melting in the hot axial conduit and the basalts from the cooler bulk of the head. More generally, we put forward stirring in plumes as one plausible mechanism contributing to compositional heterogeneity in hotspot melts. The predicted diameter of plume heads originating at the core-mantle boundary is ∼ 1000 km and this is expected to enlarge to ∼ 2000 km when the plume collapses beneath the lithosphere. This result is in excellent agreement with the observed extent of volcanism and uplift associated with continental flood volcanism. It also provides support for the hypothesis that at least some plumes originate at the core-mantle boundary.