This paper evaluates the duration of strong ground shaking that results from nuclear explosions and identifies some of the problems associated with its determination. Knowledge of the duration of horizontal ground shaking is important out to epicentral distances of about 44 km and 135 km, the approximate distances at which the ground shaking level falls to 0.01 g for nuclear explosions having yields of about 100 kt and 1,000 kt, respectively. Evaluation of the strong ground motions recorded from the event STRAIT (ML = 5.6) on a linear array of five, broad-band velocity seismographs deployed in the distance range 3.2 to 19.5 km provides information about the characteristics of the duration of ground shaking. The STRAIT data show that: (1) the definition that is used for defining duration is very important; (2) the duration of ground acceleration, as defined in terms of 90 per cent of the integral of the squared time history (Trifunac and Brady, 1975), increased from about 4 to 26 sec over the approximately 20-km distance range; and (3) the duration of ground velocity and displacement were slightly greater because of the effect of the alluvium layer on the propagating surface waves. Data from other events (e.g., MILROW, CANNIKIN, HANDLEY, PURSE) augment the STRAIT data and show that: (1) duration of shaking is increased by frequency-dependent site effects and (2) duration of shaking, as defined by the integral of the squared time history, does not increase as rapidly with increase in yield as is indicated by other definitions of duration that are stated in terms of an amplitude threshold (e.g., bracketed duration, response envelopes). The available data suggest that the duration of ground acceleration, based on the integral definition, varies from about 4 to 40 sec for a 100-kt range explosion and from about 4 to 105 sec for a megaton range explosion in the epicentral distance range of 0 to 44 km and 0 to 135 km, respectively.