Data from 110 on-farm vegetable trials conducted in Jamaica were examined to seek guidelines as to how on-farm trials should be executed in order to obtain quality data. The results show that while one-way blocking is useful in designing experiments it may sometimes be inefficient and that two-way blocking is generally preferable. If only a small number of trials can be conducted then it is important that up to four or five replications of each treatment are used at each site. Plot sizes should be larger than those used for on-station trials. Due to low correlations between results from different sites, experiments should be conducted at many sites. Thus a large number of trials should be planned to allow for treatment–environment interactions and trial failures.