Colour management systems need to be accurate and repeatable, particularly in graphic arts applications such as converting image data to proofing systems and to final print processes. Modelling and computation errors were analyzed for a number of characterization data sets including single press sheet measurements and reference printing conditions based on SWOP and ISO 12647-3. Errors were calculated at double precision and compared with the AToB, BToA and round trip errors found in ICC profiles generated from the same data sets. Data sets which are averaged from measurements of multiple press sheets consistently performed better than transforms built from single-sheet measurements. A significant part of the transform error lies in the limited precision of computation. BToA transform errors averaging 1 ΔE* ab appear to be a reasonable expectation for well-formed data sets. This needs to be placed in the context of errors arising from measurement uncertainty and the variation in colour reproduction systems, which are considerably larger.