Ð We test how well low-magnitude (m bLg 1.8 to 2.6), 25-ton chemical explosions at Balapan, Kazakhstan, can be located using IMS stations and standard earth models, relying on precisely determined relative arrival times of nearly similar, regional and teleseismic waveforms. Three 1997 Balapan explosions were recorded by a number of currently reporting and surrogate IMS stations. Three regional stations and two teleseismic arrays yielded consistent waveforms appropriate for relative picking. Master-event locations based on the AK135 model and ground-truth information from the ®rst, shallowest and bestrecorded explosion, fell under 1 km from known locations, for depths constrained to that of the master event. The resulting 90% con®dence ellipses covered 12±13 km 2 and contained the true locations; however, results for depth constrained to true depth were slightly less satisfactory. From predictions based on ground truth, we found a P g -coda phase at Makanchi, Kazakhstan to be misidenti®ed and poorly modeled. After accounting for this, 90% ellipses shrank to 2±3 km 2 and true-depth mislocation vectors became more consistent with con®dence-ellipse orientations. These results suggest that a high level of precision could be provided by a tripartite array of calibration shots in cases where models are poorly known. We hope that the successful relocation of these small Balapan shots will support the role of calibration explosions in veri®cation monitoring and special event studies, including on-site inspection.