Several thousand traditional upland rice varieties have been collected in Laos. Between 1992 and 2000, approximately 2000 of these were evaluated for grain yield and other agronomic characteristics in researcher-managed trials in northern Laos, in an effort to identify high-yielding traditional varieties for dissemination to upland farmers. A few of the tested varieties were given to farmers to evaluate in collaboration with district extension officers before 2001, but data were returned in only 6% of the onfarm trials. In 2001, a new participatory varietal selection (PVS) programme was designed and implemented to obtain information about farmer preference and on-farm performance of the upland varieties, and to increase the efficiency of the screening effort. In the first stage, on-farm PVS was conducted with 32 farmers in five provinces. Farmers evaluated eight early or eight mid-duration varieties selected by researchers in the earlier agronomic testing programme. Preference data were successfully obtained from 84% of the farms and yield data from 63%, a 10-fold increase in the on-farm trial success rate over the previous programme. Farmers strongly preferred early-maturing, large-seeded varieties with large panicles and strong stems, and disliked varieties with few tillers or tillers that ripened non-uniformly. These criteria will be incorporated into the initial stages of varietal screening in the future. The correlation between variety means over farms for grain yield and preference rating was 0.82 in the medium maturity trial and 0.54 in the early trial. Occasionally, high-yielding cultivars were not preferred, but the lowest-yielding cultivars were never preferred, indicating that agronomic selection for grain yield helps to select varieties farmers prefer. A farmer-preferred, early-maturing variety (Nok) that significantly outyielded the local checks was identified. This and five other lines are being evaluated in a scaled-up PVS programme in 2002.