A data processing technique based on impulse responses from multivariable time series was applied to radon data. It was found that soil gas radon at 50 cm depth was affected more by meteorological parameters than at 100 cm depth. This fact together with the fact that the radon change at the latter depth was larger when an earthquake occurred, radon measurement at 100 cm depth was tested as an earthquake precursor. The result showed correlation between radon anomalies and earthquake events. ã 1999 Elsevier Science Ltd. All rights reserved.