Abstract In Southeast Asia, upland rice (Oryza sativa) is an annual crop typically grown for subsistence on hilly lands that are at risk for soil erosion. If perennial cultivars of upland rice were developed, they would provide farmers with an inexpensive tool to produce a preferred food while conserving soil. O. rufipogon, the undomesticated progenitor of O. sativa, includes perennial and stoloniferous forms. To evaluate the feasibility of developing perennial cultivars by combining genes of O. sativa and O. rufipogon, four trials were conducted in an upland field at IRRI: O. sativa/O. rufipogon F1 clone and cutting height trial, F2 family trial, and two O. sativa/stoloniferous-selection factorial mating design trials. Plants were established at the beginning of the rainy season, then subjected to a 6-month dry season and evaluated for survival after 1 year. Of the 2101 cultivar control plants, only three IR47686-1-4-B individuals survived and none produced stolons. The 18 F1 clones, which were selected from previous trials with less drought stress, ranged in survival from 4.4 to 91.4%. Cutting height at harvest did not affect survival. Survival among the F2 families ranged from 9.4 to 31.9%. Segregation for stolon presence did not differ from a 3:1 ratio for five of the six F2 families, suggesting the effect of a single dominant gene. Average yields per plant for the F2 families were 1/3 to 1/9 of yields for the cultivars. However, by crossing the F1s to cultivars, yield potential was almost fully recovered. For the full-sib families of the factorial trials, survival ranged from 0.0 to 48.6%. Azucena and IR47686-1-4-B, both japonica cultivars, exhibited greater general combining ability for survival relative to the six other cultivar parents. Thus, the process of developing perennial cultivars of rice should include screening annual cultivars to identify those with the best combining ability for survival. To develop cultivars of perennial upland rice, drought avoidance and/or tolerance from annual upland cultivars must be combined with the capacity for perennial growth from wild perennial species. The frequency of stolonifereous testcross progeny was lower than expected and was affected by the O. sativa parent. Thus, additional genes likely affected stolon penetrance and expression. Analyses of covariance indicated that stolons improved the likelihood of survival for progenies of the factorial trials but the effect was small (b