We present a method for corresponding and combining cartilage thickness readings from a population of patients using the underlying bone structure as a reference. Knee joint femoral bone and cartilage surfaces are constructed from a set of parallel slice segmentations of MR scans. Correspondence points across a population of bone surfaces are defined and refined by minimising an objective function based on the Minimum Description Length of the resulting statistical shape model. The optimised bone model defines a set of corresponding locations from which 3D measurements of the cartilage thickness can be taken and combined for a population of patients. Results are presented for a small group of patients demonstrating the feasibility and potential of the approach as a means of detecting sub-millimetre cartilage thickness changes due to disease progression.