Crucial to our understanding of melt migration and segregation from source rocks beneath mid-ocean ridges and volcanic provinces is the intimate relationship with ductile mantle deformation. For this reason, numerous experiments and theoretical studies have been performed to simulate the process of melt extraction from deforming mantle rocks, with the emerging view that segregation occurs into melt-rich bands that are inclined relative to the direction of deformation. This is seemingly in contrast, however, to observations made on mantle rocks exposed at the Earth’s surface, which display evidence for melt migration in bands that were originally parallel to the direction of deformation. Here, we present experimental evidence that reconciles these contradictory observations. Olivine aggregates containing 10 wt% of a melt that reacts with olivine to precipitate orthopyroxene were deformed at 2 GPa and 1150 °C in simple shear experiments. In agreement with previous studies, we observe the development of a melt-preferred orientation that is inclined (∼30°) with respect to the main compression axis σ1. However, the alignment of newly crystallized orthopyroxene aggregates defines a fabric that develops perpendicular to σ1 and rotates toward the shear direction with increasing shear strain (γ of 0.3–2). This misalignment significantly changes the interpretation of evidence for melt channeling and transport in exposed upper mantle rocks: the fabric formed by the phases that have crystallized from a melt during deformation cannot be used directly as a marker for the melt migration direction.