Here, the implementation of the generalized F detector (Selby, 2008) as an automatic teleseismic signal detector is described. The method is applied to 10 days of waveform data from 13 small-aperture arrays of the primary seismic network of the International Monitoring System (IMS), which is being set up to monitor compliance with the Comprehensive Nuclear-Test-Ban Treaty (CTBT). The results demonstrate that the generalized F method can be used as a signal detector at small-aperture arrays, despite the correlated noise that prevents the original F detector being useful at such arrays. By comparing lists of detections with predicted arrival times of first-arriving P-type phases from earthquakes in the Reviewed Event Bulletin (REB, produced by the International Data Centre, IDC, being set up to monitor the CTBT), it is shown here that the generalized F detector compares favorably with traditional signal detection methods used at the IDC, with more candidate associations being made despite the total number of F detections being about half of those made by the IDC. Increasing the proportion of associated detections should improve the efficiency with which automatic bulletins can be constructed, and reduce analyst workload. Further advantages over traditional methods are that the F detector approach is simple to explain, based on probability theory and physical models of noise and signals, treats each array equally and objectively, and does not require subjective tuning. Analysis of the vector slowness distribution of associated and unassociated detections shows that associated detections have similar distributions for both F and the IDC, whereas the distribution of unassociated detections can be different, with unassociated IDC detections at some arrays apparently being associated with correlated noise sources.