Two sets of field experiments were conducted at two sites (lowland and intermediate altitude) for 2 years in eastern Ethiopia to determine the influence of parthenium weed (Parthenium hysterophorus) density and duration of competition on grain yield of sorghum. In the first set of experiments, target weed densities of 0, 3, 7, 13, 27, 53 and 100 plants m 72 were used. Yield loss was severely affected by parthenium weed density, peaking at 97% at the lowland site in 2000. At this site, even very low density (e. g. three plants m 72) resulted in a high yield loss (69%). Owing to differences between sites and years, however, it was not possible to specify meaningfully the threshold densities for weeding. In the critical weed-free period experiments, four parthenium weed-infested and four weed-free periods from emergence to harvesting of sorghum were maintained by hand hoeing. The critical periods for weed control, i.e. the period over which weeding had the greatest benefit on yield, were 19-69 and 40-57 days from emergence of sorghum in 1999 and 2000, respectively, at the intermediate altitude (assuming an acceptable loss of 10%). In the lowland, however, it ranged directly from emergence to 61 and 66 days, indicating more severe competition at this site. The substantial variation in yield and yield loss between sites and years illustrates the problems of attempting to give well-established, accurate recommendations for threshold densities and critical periods in small-scale, rain-fed agriculture.