Quartz crystallographic preferred orientations (CPOs) in natural mylonites can vary to such an extent that they apparently give opposite senses of shear within a single thin section. Many qualitative explanations have been proposed. Here, we take a multiscale numerical approach to investigate the variation of quartz CPOs resulting from flow partitioning in a ductile shear zone environment. We couple our self-consistent Eshelby formalism for power-law viscous composite materials with the visco-plastic self-consistent (VPSC) model for simulating CPOs in crystalline aggregates. In a quartz-bearing polyphase mylonite, we regard quartz aggregates in the rock as microscale Eshelby inhomogeneities embedded in a macroscale medium (the polyphase continuum). The effective rheology of the continuum is represented by a hypothetical homogeneous equivalent medium and is obtained self-consistently from the constituent phases (e.g., quartz, feldspar and mica grains). We obtain the partitioned flow fields in each quartz aggregate first, using our own Eshelby formalism, and then use the partitioned fields to simulate quartz CPOs, using the VPSC model. We reproduced the observed quartz CPOs and found out that the CPO variation actually reflects a macroscale finite strain gradient rather than vorticity-sense reversal as previously thought. We demonstrated that, despite the microscale flow fields varying from one quartz aggregate to another and from the macroscale flow, the sense of vorticity in all quartz RDEs remains the same as the macroscale vorticity. Our work suggests that quartz c-axis fabrics cannot be effectively used to estimate the macroscale flow vorticity.