The use of seismic anisotropy to image upper mantle flow is usually based on the assumption that the direction of fast propagation of seismic waves is the same as the flow direction. Laboratory experiments showed that when melt is present, it is rather the direction of slow propagation that aligns with the flow direction. This paper presents a modeling of the effect of melt on the development of lattice preferred orientation (LPO) and its implications for the interpretation of anisotropy. Taking into account strain partitioning associated with melt segregation in the modeling yields a good agreement with experimental results. Implementing strain partitioning in a test flow shows that the effect of melt is mainly local and has a limited effect on seismic anisotropy. Water-induced texture, melt-filled cracks or channelized flow, might better account for seismic observations of direction of fast propagation normal to plate motion on a regional scale.