The dynamical interaction between three-dimensional (3-D) buoyant flow and plate-driven mantle flow beneath a mid-ocean ridge is examined using a combination of laboratory and numerical experiments. In a unique laboratory setup a layer of strongly temperature-dependent viscous fluid is heated from below and cooled from above to drive thermal convection. Forced, plate-driven flow is modeled by dragging mylar sheeting in opposite directions across the fluid surface. Uniform viscosity 3-D numerical models have been designed to simulate the laboratory runs, to provide additional information on the flow, and to attempt to isolate the effects of variable viscosity. In one type of experiment we modeled buoyancy on the scale of the partial melting region with a linear heat source beneath the spreading axis. In a second set of experiments the entire base of the tank is heated in order to investigate the interaction between plate-driven flow and larger-scale (upper mantle) convection. The pattern of segment-scale (~100--150 km) convection beneath a spreading center is found to be a strong function of the spreading rate and the Rayleigh number (Ra) of the buoyant flow. Purely two-dimensional (2-D) flow exists only in the case of low Ra (~105) and faster spreading rates (>4--6 cm/yr half rate). Three dimensionality is strongly enhanced during transient pulses of upwelling. Convection on the scale of the upper mantle can contribute significant long-wavelength spatial (~600--1000 km) and temporal (20--40 m.y.) variability in upwelling rates and temperatures at spreading centers and may provide an alternative model for plume-ridge interaction. For a given Ra the upwelling near the spreading axis can essentially be described by three regimes: weakly 3-D at the slow spreading rates, strongly 3-D at slow to intermediate rates, and 2-D at fast spreading rates. The regime boundaries shift toward higher plate velocities with increasing Ra. Comparison of the laboratory and numerical experiments indicates that temperature-dependent viscosity may strengthen the position of focused upwelling centers and cause a sharper transition between 2-D and 3-D upwelling patterns. Taken together, these results suggest that buoyancy-driven, dynamic flow is an important element in the geodynamics of mid-ocean ridges. ¿ American Geophysical Union 1996