Excessive femoral varus has been implicated in the aetiology of canine medial patellar luxation. Several reports have demonstrated that malpositioning for radiography in terms of either femoral elevation or rotation can have marked effects on the measured varus angle. Three-dimensional reconstruction of CT images offers the clinician the possibility to eliminate these positional confounders. Although one recent study reported good repeatability for measurements made from a single reconstruction, we are not aware of previous reports of reproducibility of varus measurements between reconstructions, a situation with more real-life applicability. CT scans of 20 canine femora underwent 3D reconstruction by 3 independent observers. Reconstruction spin and tilt data were used to assess reconstruction variability. Two observers of differing experience levels made 3 independent readings of their own reconstructions, and measurement reliability was assessed using the repeatability coefficient. The repeatability coefficient represents the interval within which two measurements on the same patient are expected to fall in 95% of cases. Coordinate data from one set of reconstructions and the spin and tilt data obtained above were used to specify a model to predict the effect of reconstruction variability on varus measurements. Intraobserver repeatability coefficients were 2.4° and 2.6°, and the interobserver repeatability coefficient was 3.5°. Reconstruction variability yielded a spin-tilt ellipse area of 0.59 deg². Surprisingly, reconstruction variability produced minimal effects on simulated varus measurements in contrast to previous experimental reports. Possible explanations include changing landmark appearance which cannot be modelled and lower magnitude of femoral subtense (procurvatum) in our specimens. CT varus measurements can be performed with clinically relevant levels of precision, regardless of experience level. Provided strict reconstruction criteria are followed, reconstruction variability will be low and likely to have little influence on the values obtained.