The role of assimilation at the roof of the axial magma chamber (AMC) at fast spreading mid-ocean ridges is investigated using field, geophysical, and geochemical data. Field observations from ophiolites indicate that roof assimilation is a common process, but provide little constraint on the amount of material assimilated. Arguments based on geophysical data that include the subsidence of the layer 2A/2B boundary across the width of the AMC, and the depth of large axial summit troughs, require at least ~50--100 m of assimilation. However, significant variations in the depth of the AMC on the East Pacific Rise (EPR), over short along-axis distances, suggest up to ~500 m fluctuations in the depth of the AMC may be common on short timescales. Geochemical modeling of the over-enrichment of chlorine in EPR basalts suggests that ~20% of the oceanic crust may go through a cycle of crystallization, alteration, then assimilation. This is far more than can be accounted for by roof subsidence and suggests that fluctuations in the level of the roof must account for the majority of the assimilation. Assimilation most likely occurs on a number of different timescales ranging from decadal, due to perturbations in the thermal regime related to diking, to longer-period fluctuations controlled by magma supply from the mantle. Large-scale roof assimilation has implications for the accretion of the lower crust due to heat losses associated with assimilation, and for the interpretation of the geochemistry of the oceanic crust.