Discrete samples of geothermal water have been collected from a network of nine stations for analysis of radon (^(222)Rn) content. The sampling network consisted of wells that range in wellhead temperature from 48°C to 100°C, and the depths range from 38 m to 1338 m. The sampling frequency at most stations was about once per week and twice per week at the station Fludir. The wells are either artesian or pumped more or less continuously. The network covered two regions of transform faulting in Iceland with seven stations in the Southern Iceland Seismic Zone (SISZ) and two stations in northern Iceland in the Tjornes Fracture Zone (TFZ). During 1978 and 1979 several anomalous changes in radon content were observed to precede some of the local earthquakes. Criteria based on tectonic regimes, a magnitude-distance relationship and time clustering were applied to select a set of 23 earthquakes that could be expected to be preceded by a radon anomaly. The magnitude of these earthquakes ranged between 1.0 and 4.3. Each of the 23 earthquakes was within the distance range of one or more stations such that altogether 57 potential observations of possible anomalies were available. The method of analysis applied to the radon and earthquake data consisted of identifying radon anomalies in retrospect, and resulted in nine precursory anomalies, 48 cases of failure to observe an anomaly, and seven false alarms. The probability of observing radon anomalies before earthquakes with magnitudes between 2.0 and 4.3 (the largest event observed) was found to be approximately 65% based on a weekly sampling rate. In the SISZ, five out of eight earthquakes (M > 2) were preceded by an anomaly. In two cases, anomalies were observed at two different stations prior to the same earthquake. The anomalies appeared to occur farther away for larger earthquake magnitude. An aquiclude that divides the SISZ did not seem to affect the occurrence of radon anomalies, but the amplitude of anomalies on the east side were larger than the ones on the west side. The duration times of the anomalies ranged from 17 to 37 days. The large number of failures to observe an anomaly indicated that the occurrence of an anomaly was more strongly dependent on local conditions and earthquake parameters than distance between epicenter and radon station. Some stations also appeared to be less sensitive than others and three stations never showed an anomaly. During periods of bursts in local seismicity, some stations reflected distinctively larger background fluctuations, which hampered the correlation of the seismicity with the radon data. Most of the false alarms were related to disturbances caused by changes in well operation. In the SISZ the high rate of failure to observe an anomaly and the occurrence of false alarms was partly compensated for by operating a dense network of sampling stations. Wells that are utilized at moderate flow rates are less likely to show false alarms than wells utilized at high flow rates. Wells situated close to rock formations rich in radioactive minerals may be more likely to show anomalies than others.