We present precise measurements of the resistance-temperature variation of several samples of neutron transmutation doped (NTD) germanium, at temperatures from 70 mK to 1 K. This material is widely used for sensitive thermometry, often as the thermistor element in bolometers and microcalorimeters. It is also used in investigations of the low temperature conductivity of highly doped semiconductors. The resistance, R, is expected to follow the variable range hopping equation R(T)=R 0 exp(T 0/T) p , where T is temperature and R 0 and T 0 are material parameters. A value of p=0.5 is predicted theoretically, and generally seems to be in good agreement with experimental measurements. However, some theories and numerical calculations predict different values of p. Knowledge of the correct resistance-temperature relationship is important for the accurate calibration of thermometers, and also delivers insight into the basic physics involved. Most experimental measurements on germanium have not had sufficient precision to distinguish between the different predicted values of p. We show that such measurements are nevertheless possible. Our results are all in excellent agreement with the expected variable range hopping behaviour. However, the values of p appear to vary with doping density, in disagreement with most theories. We have considered and rejected both random and systematic errors as an explanation for the observed behaviour, and have confirmed the results by making measurements in two different systems with independent readout systems and temperature calibrations. The situation is complicated by the possibility of temperature dependence of R 0. The expected form is R 0(T)∝T q; however, there is considerable disagreement over the predicted value of q. We show that in general it is not possible to determine both p and q from resistance measurements. However, our results can only be explained if either or both of q and p vary from sample to sample. Such behaviour is not generally expected. We show that neglecting the q term can lead to serious errors when calibrating thermometers. However, the degeneracy between p and q means that for a calibration the q term can be neglected, and good fits obtained if p is allowed to vary. Our results suggest that further theoretical work is required in this area, backed up by more comprehensive measurements.