A large number of theoretical, heuristic and empirical models for transport in a Tokamak have been developed. Many models have been claimed to adequately describe subsets of the wealth of data accumulated from several Tokamak experiments. Yet no single transport model has emerged as an obvious candidate that will give an accurate extrapolation from today's large Tokamaks to a reactor. It is shown how this situation can arise from three features associated with the testing of models against experimental data: i) measurement errors, ii) collinearities in the data and iii) confinement degradation with power. These features, when combined, permit the claims that a variety of transport models based on different physics can fit the data. It is proposed that transport models should be tested against data from experiments in which only one dimensionless plasma parameter is being varied at a time. Such experiments include the scans carried out on DIIID, TFTR and JET of the parameters r* (normalised Larmor radius), n* (collisionality) and b. It is also shown that the testing of transport models must include data from two or more Tokamaks of different size in order to separate model predictions by more than the experimental measurement error level.