Upland rice (Oryza sativa L.) yields on infertile, acid soils in the Asian uplands average only 1 Mg ha-1. Phosphorus deficiency is considered a major soil constraint to increased yield, but little quantitative information is available. We analyzed P responses of traditional rice on farm in Laos, Thailand, and the Philippines and improved varieties in researcher-managed trials in the Philippines. Treatments in onfarm trials were a control and 50 kg P ha-1 ± 100 and 50 kg ha-1 N and K, respectively. Treatments in researcher-managed trials were P rates on an unlimed and limed Ultisol. Mehlich-1 extractable P correlated with on-farm grain yield (r = 0.47). Phosphorus fertilization increased average grain yield (1 Mg ha-1) by 20%, total biomass (4 Mg ha-1) by 27%, and P uptake (4.1 kg ha-) by 53%. Yield increased 37% with P + N + K, but only 16% of the 2.4 Mg ha-1 biomass increase was grain. Improved rice in researcher-managed trials responded to P, with a larger proportion of biomass partitioned to grain [i.e., higher harvest index (HI)]. Grain yield of 'UPLRi-5' increased from 3.2 to 4.6 Mg ha -1 in limed soil while that of 'IR55423-01' increased from 3.4 to 4.0 Mg ha-1 in unlimed soil. Phosphorus fertilization always increased the frequency of higher yields averaged across trials, soils, varieties, and growing conditions. Yield gain from onfarm P fertilization of traditional rice was small because of low HI, unlike in improved varieties, which had a HI. We infer that increasing upland rice yield in Asia would require genotypes with higher HI in addition to P fertilization.