Dike intrusion is a fundamental process during upper oceanic crustal accretion at fast- to superfast-spreading ridges. Based on the distribution of magma along fast-spreading centers inferred from marine geophysical data, models predict systematic steep flow at magmatically robust segment centers and shallow magma flow toward distal segment ends. Anisotropy of magnetic susceptibility (AMS) fabrics from 48 fully-oriented block samples of dikes from upper oceanic crust exposed at Hess Deep Rift and Pito Deep Rift reveal a wide range of magma flow directions that are not consistent with such simple magma supply models. The AMS is interpreted to arise from distribution anisotropy of titanomagnetite crystals based on weak shape-preferred orientation of opaque oxide and plagioclase crystals generally parallel to AMS maximum eigenvectors. Most dike samples show normal AMS fabrics with maximum eigenvector directions ranging from subvertical to subhorizontal. The distributions of inferred magma flow lineations from maximum eigenvectors show no preferred flow pattern, even after structural correction. We use a Kolmogorov Smirnov test (KS-test) to show that the distribution of bootstrapped flow lineation rakes from Pito Deep are not statistically distinct from Hess Deep, and neither are distinguishable from Oman and Troodos Ophiolite AMS data. Magma flow directions in sheeted dikes from these two seafloor escarpments also do not correlate with available geochemistry in any systematic way as previously predicted. These results indicate distinct compositional sources feed melt that is injected into dikes at fast- to superfast-spreading ridges with no preference for subhorizontal or subvertical magma flow. Collectively, results imply ephemeral melt lenses at different along-axis locations within the continuous axial magma chamber and either direct injection or intermingling of melt from other deeper ridge-centered or off-axis sources. (C) 2014 Elsevier B.V. All rights reserved.