We analyze over 1300 km of high resolution along-axis gravity profiles at ridges with half-spreading rates ranging from 1.2 to 5.5 cm/yr. The results show consistently higher along-axis gradients of mantle Bouguer anomaly at the slow-spreading Mid-Atlantic Ridge (MAR) (0.3-1.2 mgal/km) than at the intermediate- to fast-spreading Cocos-Nazca Ridge and East Pacific Rise (EPR) (0.1-0.2 mgal/km). The regional peak-to-trough amplitude of mantle Bouguer anomaly is also greater along the MAR (30-60 mgal) than the Cocos-Nazca Ridge and the EPR (10-20 mgal). With increasing spreading rate, the regional peak-to-trough amplitude of axial seafloor depth decreases from 1000-1700 m to 200-700 m. 3-D numerical experiments suggest that mantle contributions to the gravity can be significant only near large-offset transforms. At the more commonly observed non-transform offsets, gravity anomalies will reflect crustal thickness variations. The along-axis gravity data thus indicate that the amplitude of along-axis crustal thickness variation decreases with increasing spreading rate. We propose that this spreading rate dependent crustal accretion style may originate in the mantle: finite-amplitude mantle upwelling is intrinsically plume-like (3-D) beneath a slow-spreading ridge but more sheet-like (2-D) beneath a fast-spreading ridge. Such a transition in mantle upwelling may occur if the relative importance of passive upwelling over buoyant upwelling increases with increasing spreading rate. Small amplitude 3-D upwellings may occur at a fast-spreading ridge, but their effects on crustal thickness variations will be significantly reduced by along-axis melt flows along a persistent low-viscosity crustal magma chamber. In contrast, the large crustal thickness variations due to 3-D mantle upwellings will be maintained at a slow-spreading ridge because less along-axis melt flows can occur in the colder and more rigid crust there. ¿American Geophysical Union 1992