A regional prediction experiment on the distribution of the atmospheric ozone column over China is performed by means of a spatiotemporal artificial neural network system (STANNS) built by using the monthly mean values of Total Ozone Mapping Spectrometer (TOMS) data from the Nimbus 7 satellite. The experiment region is situated from 72.5 ¿E to 128.75 ¿E and from 22 ¿N to 49 ¿N and divided into a rectangular grid net of 16¿10. The data used cover 174 months from November 1978 to April 1993; those from the first 150 months were used to construct STANNS, and those from the last 24 months were used to test the prediction ability. The experiment results are encouraging. All the values of the correlation coefficient between the prediction and the actual fields for 24 or more samples exceed 90% at the lead time of not more than 6 months. This shows that there exists a good agreement between the prediction and the actual fields. In addition, some important features in the ozone distribution, such as the unusual depletion area over the Tibetan Plateau in summer, are also predicted very well. The results also show that, in a rough sense, the prediction accuracy decreases with the increasing of lead time. ¿ 2000 American Geophysical Union