At a depth of ~660 km in the Earth's mantle the spinel-perovskite phase boundary is a prominent barrier for mantle convection. This is due to the negative Clapeyron slope of the phase equilibrium curve which leads to an elevation of the phase boundary within hot upwellings causing negative buoyancy forces. We have investigated the conditions for rising plumes either to penetrate and pass the spinel-perovskite phase boundary or to stick and spread below it by studying the fundamental physics of this process. The plume heads were simply modeled as hot three-dimensional (3-D) spheres or 2-D cylinders. A simple calculation balancing the positive thermal and the negative phase buoyancy forces leads to a better parameterization using two dimensionless quantities. In addition to the phase buoyancy parameter, we defined a deflection parameter, relating the elevation of the phase boundary to the plume head radius to account for the geometrical shape of a plume head. This parameterization is further tested with numerical models that include the effects of thermal diffusion, latent heat, the olivine-spinel phase boundary at a depth of 410 km, and temperature and/or phase-dependent viscosity structure. For laboratory estimates of the slope (-3 MPa/K) and density increase at the spinel-perovskite phase boundary (250 kg/m3) our models predict that plumes with excess temperatures of 50¿--600 ¿C will stick at the top of the lower mantle if their head radii are less than ~100 km. Plumes will penetrate into the upper mantle if plume head radii exceed 100 km. While the style of plume penetration or spreading at the top of the lower mantle strongly depends on the viscosity structure, the conditions for penetration do not. All rising hot volumes with nonpenetrating conditions stick at the top of the lower mantle and spread laterally, independent of their viscosity structure. For weakly nonpenetrating conditions, heat diffusion increases the radius of the hot volume and leads to penetration during a secondary stage. For strongly nonpenetrating conditions, spreading at the top of the lower mantle drives a mechanically coupled counterflow in the upper mantle, which is stable for a very long time. For volumes with an excess temperature of more than 350 ¿C heat diffusion across the phase boundary will eventually inhibit this counterflow and stabilize a thermally coupled flow which might entrain some material of the hot volume. However, our results suggest that the spinel-perovskite phase boundary is unlikely to inhibit the penetration of mantle plumes of the size thought to have generated many flood basalt provinces and hotspot chains. ¿ 2000 American Geophysical Union