Summary. Why primary school children have considerable difficulty in measuring and estimating temperature is not known. It may be because they do not understand temperature or more simply that they do not understand the temperature scale. The latter is suggested by the improvement in their temperature predictions when the temperatures are described verbally rather than numerically. However, success on verbal versions may be an artefact of the procedures used. Two experiments comparing numerical and verbal temperature predictions are reported: one involving children between 9 and 12 years old, the other with children between 7 and 12. In both studies verbal temperature predictions were markedly superior but few children were consistent in predicting the effects of combining and dividing quantities of water. The second study showed that confusion over temperature was just as common among younger children and therefore unlikely to be simply a product of integrating faulty quantitative temperature beliefs with sound intuitions.