This study reports on variations in seismic activity preceding moderate and large earthquakes in a highly active portion of the India/Eurasia collision zone. The study is based on the catalog of earthquake locations from the Complex Seismic Expedition (CSE) network in the Garm region, Central Asia. The local seismic network was established in the mid-1950s and recorded and located over 85,000 earthquakes from 1955 to 1988. Two main approaches were used to evaluate variations in seismic activity associated with 12 moderate- to large-magnitude (ML = 5.0 to 6.3) mainshocks that occurred in the study area: (1) a qualitative approach, involving examination of maps, cross sections, space-time plots, and earthquake histograms for the area identified as the rupture area of the mainshock and (2) a quantitative approach based on magnitude-frequency relations estimated for small space-time windows within the seismic network. In addition to temporal variations in seismicity, three long-term gaps were identified: (1) in the western Peter the First Range, near the city of Garm; (2) in the central Peter the First Range; and (3) in the northeastern part of the study area, approximately coinciding with a gap identified by other authors using different approaches. The potential magnitude of future gap-filling events is estimated to be in the range 6.7 to 7.0. Evidence of precursors has been found to varying extent for 9 out of the 12 mainshocks studied. These include quiescence or low activity before the mainshock, foreshock activity before the mainshock, and a sequence of low-activity-high-activity-quiescence before the mainshock. For the quantitative analysis, the level of seismic activity in small space-time windows is quantified by the parameter A2, defined as the A value at magnitude 2 in the Gutenberg-Richter magnitude-frequency relation log N = A - bM. The A2 values were calculated for 30 12 × 12-minute spatial cells with a moving time window of five years. From this analysis, only one earthquake, the M = 6.3 Dzhirgital earthquake of 26 October 1984, shows a significant decrease (by a factor of 45%) in activity level A2. The results of the quantitative analysis also show strong spatial variations: the central portion of the study area, which includes the Peter the First Range fold-thrust belt, shows high, but relatively stable, seismic activity; the Tien Shan basement uplift in the north has a tendency of decreasing activity; and the northeastern part of the study area shows increasing activity. The October 1984 earthquake is located near the transition between the two tendencies.