In contrast to the along-axis uniformity observed at the East Pacific Rise (EPR), crustal accretion at the Mid-Atlantic Ridge (MAR) appears to be a highly complex and heterogeneous process. Besides spreading rate, one of the first-order differences between the EPR and the MAR is the much higher degree of ridge segmentation observed in the Atlantic. Circular lows in the mantle Bouguer anomaly (MBA bull's-eyes) are common at centers of spreading segments of the MAR, suggesting crustal thickness variations of up to 4 km along individual segments. We use a three-dimensional numerical model of mantle flow to examine the effect of ridge segmentation on mantle upwelling and the resulting overall crustal production and along-axis variations in crustal thickness. Mantle flow in our model is driven by both buoyant forces and segmented plate spreading. Various asthenospheric viscosity structures, plate spreading geometries, and mantle potential temperatures are explored. We find that a combination of buoyant mantle flow and three-dimensional melt migration can reproduce crustal thickness variations similar to those inferred from gravity. Buoyant flow gives rise to variations in upwelling velocity at along-axis wavelengths greater than 150 km but does not contribute to short-wavelength variations. However, three-dimensional melt migration may greatly enhance crustal thickness variations along all segments, independent of the wavelength of buoyant upwelling. We present an idealized model, in which melt first rises vertically and then flows along the base of the lithosphere toward the ridge axis, that easily produces crustal thickness variations greater than 4 km. The models also predict that the average crustal thickness should decrease with increasing amount of segmentation and decreasing spreading rate. Therefore the thinner, more heterogeneous crust observed at the MAR may result from the combined effects of slower spreading rate and more pervasive ridge segmentation.¿ 1997 American Geophysical Union