The statistical sampling theory given by Lumley and Panofsky (1964) can be used to compute the temporal sampling interval and the total sampling duration required for statistical studies of the long-term horizontal dispersion of pollutants. To illustrate the technique, the release of a series of tracer particles from St. Louis has been simulated, and trajectories have been computed from observed winds. An analysis of trajectories according to season indicates that the minimum interval required for statistical independence is 35 hours. To limit the statistical error (&egr;) due to insuffcient averaging time to 10%, the corresponding required durations of sampling are 1300 and 2900 hours for winter and spring, respectively. If the year is examined as a whole, a sampling duration of 2400 hours is required. These sampling durations are highly dependent upon &egr;, and if accuracy is limited to within 30%, they may be decreased by about an order of magnitude.