The effect of grassland gap size on the establishment of five non-clonal perennial forbs, comprising a range of seed masses, was investigated in two Themeda triandra Forsskal grasslands by sowing seed in the centre of artificially created 6 cm (36 cm2) and 18 cm (324 cm2) wide canopy gaps. Undisturbed vegetation (0 cm gap) was used as a control. Initial seedling establishment and growth in gaps was measured over 24 weeks, as was that of transplanted juvenile plants, and all were related to light availability at ground level. The rate at which natural canopy gaps in grasslands disappear with time since burning was also quantified in a chronosequence study to infer how long gaps persist after disturbance by fire. For four of the five species, relatively few seedlings emerged (< 30% of sown seed germinated), and emergence was largely independent of the effects of gap size, sward structure and seed mass. Only Plantago gaudichaudii Barneoud germination was inhibited under a closed sward (0 cm gap). Hence, the presence of vegetation per se largely did not affect germination for most species. Survival of seedlings, however, was negatively affected by vegetation, presumably because of low light levels in the small gaps. Eryngium ovinum Cunn., P. gaudichaudii and Velleia paradoxa R.Br. failed to establish in 0 cm gaps in the 3-year-old sward. Gaps in the vegetation, however, enabled all species to survive at low levels in at least one of the grasslands, with survival of seedlings in 18 cm gaps usually greater than that in 6 cm gaps. Survival of transplants was initially high (> 50%) for all species in the 2-year-old sward, but E. ovinum and Senecio macrocarpus Belcher mortality was substantial in 0 cm gaps in the 3-year-old sward where light levels were lowest (i.e. < 9% of available light). For these species, small gaps are likely to be hostile microsites for establishment at all times. Growth was also affected by gap size, most species positively responding to 18 cm gaps by growing substantially bigger than transplants in 0 cm gaps, particularly in the more shaded 3-year-old sward gaps. The large canopy gaps (i.e. > 300 cm2) that enable the most effective recruitment in this grassland are rare in the years after fire and constitute only 8–19% of all gaps that are present at 1 and 2 years after burning. By 3 years after fire, canopy gaps are small (< 30 cm2) and infrequent (< 3 gaps m–2) as a result of rapid recovery of the dominant grass. These small gaps would appear to provide few opportunities for substantial seedling recruitment of many species by this time. If seedling recruitment events are to be optimised for the herbaceous dicotyledons that characterise this community, disturbance to the canopy is required at intervals of 1–3 years. This regime becomes particularly important when other features of the recruitment ecology of this grassland are considered.