One of the uses of ISO 7730 (predicted mean vote, PMV) is to predict the thermal sensations of people in buildings. This application is examined, using the ASHRAE database of field-studies. Taking these world-wide data as a single distribution, PMV is free from serious bias. There exist, however, underlying biases in relation to all contributing variables, and a further bias related to the outdoor temperature. These biases often combine to produce a substantial bias in PMV. In surveys of individual buildings, PMV often differs markedly and systematically from the actual mean vote, both for naturally ventilated (NV) and for air-conditioned (AC) spaces. Possible origins of the biases are discussed, and it is shown that it would be possible to modify PMV substantially to reduce them. Environmental consequences of the use of PMV are discussed. It is concluded that ISO 7730 in its present form can be seriously misleading when used to estimate thermal comfort conditions in buildings.