Segmentation and along-axis variations within individual segments indicate the inherently three-dimensional nature of mantle upwelling and melting beneath oceanic spreading centers. Numerical convection experiments are used to explore the effects of local buoyancy forces on upwelling and melt production beneath a segmented spreading center. The experiments are conducted in a region consisting of a thermally defined rigid lithosphere and a uniform viscosity asthenosphere overlying a higher-viscosity mantle half-space. A periodic plate boundary geometry is imposed consisting of spreading segments and transform offsets. Buoyancy forces are caused by thermal expansion and the compositional density reduction due to the extraction of partial melt. The relative magnitudes of the buoyant and plate-driven components of mantle flow are controlled by the spreading rate and mantle viscosity, with buoyant flow more important at lower spreading rates and viscosities. Buoyant flow beneath the spreading axis amplifies along-axis variations in upwelling near a ridge-transform intersection, and distributes the variations along the entire spreading axis. Buoyant flow may thus be responsible for the more three-dimensional character of slow spreading centers. Away from the spreading axis, thermal buoyancy drives convective rolls that align with the direction of plate motion and which have an along-axis wavelength controlled by the prescribed thickness of the asthenosphere. However, the position and stability of rolls are influenced by the segmentation geometry. In cases where the spreading center geometry does not allow a stable configuration of rolls, the flow is time-dependent. Along-axis variations in upwelling cause variations in melt production, which imply large variations in crustal thickness that dominate the surface gravity signal. The crustal thickness distributions implied by these numerical experiments produce bulls-eye-shaped negative mantle Bouguer anomalies centered over spreading segments, as observed at several spreading centers. The amplitude of the anomaly increases with decreasing spreading rate. ¿ American Geophysical Union 1993