Magma chambers cool from all sides, with expected vigorous convection. Strong changes in viscosity with crystallization, however, lock up within an inwardly propagating crust a great deal of buoyancy otherwise available to drive convection. With sufficiently small degrees of superheat, the Rayleigh number (Ra) is small and convection is transient and never becomes fully developed. In order to understand this transient stage of convection, we have performed some fluid dynamic experiments on the solidification of a paraffin layer. Except in the growing crust and its thin forward boundary layer, the interior remains isothermal and the temperature decreases uniformly until it is locked at the convective liquidus. The crystals are fine and hair-like dendrites and no major differentiation of the fluid composition is observed. The time-scale of the convective stage is predictable and is very short relative to the time of complete solidification. We also report measurements over time of crust thickness, convective velocity, and heat transfer. When the interior temperature reaches the convective liquidus, convection wanes, although the calculated Ra, based on an arbitrary small temperature contrast of 0.2¿C, is still large (~107). A small (¿0.2¿C) uncertainty in ΔT (i.e., 0≤ΔT≤0.2¿C) produces a large uncertainty in Ra (i.e., 0≤Ra≤107). This uncertainty in Nature may preclude an accurate determination of Ra in a magma chamber residing at a temperature very close to the convective liquidus. ¿ American Geophysical Union 1990