The thixotropy of cellulose nanocrystal (CNC) water suspensions is intrinsically dependent on the hierarchical structure of the suspension. The diverse hierarchies that comprise individual CNC nanoparticles and mesophase liquid crystalline domains, chiral nematic and nematic structures, contribute selectively to the rheological material response. Here, we combine rheology with polarized light imaging (PLI) to elucidate the thixotropic behavior of CNCs suspended in water. The simultaneous monitoring of PLI and rheological tests enables the observation of mesogens and their orientation dynamics. Creep, dynamic time sweep, ramped hysteresis loop, and thixotropic recovery tests combined with PLI aim to differentiate the contribution of the different hierarchical levels of CNC suspensions to their thixotropy. The range of concentrations investigated comprised biphasic (4 and 5 wt. %) and liquid crystalline phase suspensions (6, 7, and 8 wt. %). The CNC suspensions exhibited complex thixotropy behavior, such as viscosity bifurcations in creep tests and overshoot in ramped hysteresis loop tests. The restructuring and destructuring appeared to correspond to different levels of their hierarchical structure, depending mainly on the phase, in agreement with previous studies. Restructuring was attributed to re-organizations of an individual CNC, e.g., in the isotropic fraction of biphasic suspensions and at the mesogen interfaces in liquid crystalline phase suspensions. However, by increasing liquid crystalline fraction in the biphasic concentrations, restructuring could also involve mesogens, as indicated in the creep tests. For flow conditions above the yield stress, as evidenced by the ramped hysteresis and thixotropy recovery tests, destructuring was dominated by orientation in the flow direction, a process that is readily observable in the form of PLI “Maltese-cross” patterns. Finally, we show that a simple thixotropy model, while unable to capture the finer details of the suspension’s thixotropic behavior, could be employed to predict general features thereof.