A long-standing problem in observational seismology is that of automating network operation. In this study we report an experiment that uses the detector described in Ruud and Husebye (1992) for automatically picking P and S arrivals in local event records from the Norwegian Seismic Network (NSN). The analysis was performed on prescreened waveform segments known to contain local events. For automatic epicenter determination a novel, robust grid-search method well-suited for estimation problems with nonGaussian observational errors is introduced in order to handle outliers. In our experiment, 49 local events from the August 1991 NSN bulletin were located. The number of detecting stations varied from 3 to 10, with an average of 5 of a total of 15 stations. P- and S-picking errors were small, mostly within 0.5 sec for both P and S. Mis-identification of the P- and S-phase was more frequent; this affected about 15% of the picks. Even for events with several large arrival time outliers (up to 40 to 50% of total) satisfactory epicenter determinations were obtained. The median location difference of our “automatic” solutions compared with those in the NSN bulletin based on analyst picked and identified phases was 15 km. For 90% of the events the difference was less than 50 km.