Dispersion of backward air parcel trajectories that are initially tightly grouped near the tropical tropopause is examined using three ensemble approaches: “RANWIND,” in which different ensemble members use identical resolved wind fluctuations but different realizations of stochastic, multifractal simulations of unresolved winds; “PERTLOC,” in which members use identical resolved wind fields but initial locations are perturbed 2° in latitude and longitude; and a multimodel ensemble (“MULTIMODEL”) that uses identical initial conditions but different resolved wind fields and/or trajectory formulations. Comparisons among the approaches distinguish, to some degree, physical dispersion from that due to data uncertainty and the impacts of unresolved wind fluctuations from those of resolved variability. Dispersion rates are robust properties of trajectories near the tropical tropopause. Horizontal dispersion rates are typically ~3°/d, which is large enough to spread parcels throughout the tropics within typical tropical tropopause layer transport times (30–60 days) and underscores the importance of averaging large collections of trajectories to obtain reliable parcel source and pathway distributions. Vertical dispersion rates away from convection are ~2–3 hPa/d. Dispersion is primarily carried out by the resolved flow, and the RANWIND approach provides a plausible representation of actual trajectory dispersion rates, while PERTLOC provides a reasonable and inexpensive alternative to RANWIND. In contrast, dispersion from the MULTIMODEL calculations is important because it reflects systematic differences in resolved wind fields from different reanalysis data sets. Key Points Dispersion rates are robust properties of trajectories near the tropical tropopause Dispersion is primarily carried out by the resolved flow Physical dispersion can be distinguished from trajectory error