Dosimetry protocols recommend that ionization chambers used in radiation therapy be pre-irradiated until they 'settle', i.e., until a stable reading is obtained. Previous reports have claimed that a lack of pre-irradiation could result in errors up to several per cent. Recently, data collected for a large number of commonly used ion chambers at the Institute for National Measurement Standards, NRC, Canada, have been collated and analysed, with additional data contributed by the National Physical Laboratory, UK. With this data set, it was possible to relate patterns of ion chamber behaviour to design parameters. While several mechanisms seem to contribute to this behaviour, the most obvious correlations implicate the type of insulator surrounding the central collector electrode, the extent of collector electrode shielding and possibly the area of the insulator exposed at the base of the active air volume. The results show that ion chambers with electrode connections guarded up to the active air volume settle quickly (~9 min) and the change in response is small (less than ~0.2%). For ion chambers where the guard connection surrounding the central collector does not extend up to the active air volume, settling times of 15–20 min and an associated change in response of up to 1% are typical. For some models of ion chambers, the irradiation rate may also play a role in settling behaviour. Settling times for the ion chambers studied here were found to be independent of beam quality.