Many different models for energy transport in a tokamak have been tested against subsets of the wealth of data accumulated from tokamak experiments and claimed to describe adequately the data used in the tests. Yet no single transport model has emerged as an obvious candidate for accurate extrapolations from today's large tokamak to a reactor. The principles involved in testing a model against data are examined. It is shown how: (i) collinearities in the data, (ii) confinement degradation with power and (iii) measurement errors in the data, can make tests insensitive to particular details of transport models. This lack of sensitivity is illustrated via tests of models against global confinement data from several tokamaks (ITER L mode database) and via tests of different models against local confinement data from one single tokamak (JET database). The latter tests demonstrate that the testing of transport models must include data from two or more tokamaks of different size