The reproducibility of some thermometric fixed points and the accuracy of four platinum resistance thermometers (PRTs) were studied. It was found that the fixed points of aluminium (Al), zinc (Zn), tin (Sn), indium (In) and gallium (Ga) were realized reproducibly within ±0.17 mK; ±0.11 mK; ±0.10 mK; ±0.13 mK and ±0.12 mK, respectively. Because the actual impurities and their concentration in our samples (of 99.9999% or 99.999 99% purity) are unknown, the systematic uncertainty due to impurities cannot be estimated. However, any of the samples of Ga, In, Sn, Zn and Al is consistent with the rest within ±0.2 mK, using a cubic or quadratic deviation function, in the temperature range 0 °C to 660 °C. This indicates that the effect of impurities is negligible. Four PRTs were selected at random. They were calibrated repeatedly, first up to the Zn point and then up to the Al point. The resistance of each PRT drifted. From time to time, for each PRT, a seemingly well-established resistance drift suddenly and unpredictably changed to a different rate of drift. Occasionally, the resistance of the PRTs shifted. Such unpredictable changes obviously limit the accuracy of temperature measurements using PRTs no matter what the accuracy of their calibrations. In the case of our four PRTs, the uncertainty of temperature measurements near 660 °C ranged from about ±1 mK to about ±2.5 mK even though they were all calibrated at all fixed points well within ±0.25 mK uncertainty. Possible explanations are offered for the apparently permanent drifts and the erratic shifts in the resistance of the PRTs. Some comments are made concerning the ambiguity of "immersion tests" in general. The furnaces of the National Research Council of Canada used in this work are high-temperature adiabatic calorimeters.