Methods and systems for aligning head scans of a subject for a virtual avatar can be based on locating eyes of the subject in the scans. After one or more eyeball models are fitted to reference candidate points of a sclera of each eyeball of the subject in a reference head scan, an additional reference point can be inferred from the eyeball models. The eyeball models can be fitted to candidate points of the sclera of each eyeball of the subject in another head scan and an additional point can be inferred from the fitted eyeball models. An affine transformation can be determined between the head scans based on the eyeball models fitted to the candidate points in the reference head scan and the other head scan and the additional points inferred. The methods and systems can be used for rigging or animating the virtual avatar.