The effect of two-component rice cultivar mixtures on the control of rice blast disease was studied in three different experiments under rainfed upland conditions in the Madagascar Highlands. The mixtures involved a susceptible cultivar (either susceptible or very susceptible) and a resistant cultivar in different mixture arrangements (random or row mixtures) and with different proportions of the susceptible cultivar (50, 20 and 16·7%), which were compared to the susceptible cultivar grown in a pure stand. The effect of these mixtures on the incidence and severity of leaf and panicle blast was measured weekly, and on yield and yield components at harvest time. The mixture effect was more efficient in reducing disease with a proportion of 16·7% susceptible component than with a proportion of 50%. Blast epidemic was significantly reduced in all three experiments. However, under high blast pressure, there was no reduction in the disease by the end of the epidemic and yields of the susceptible cultivar were almost zero whatever the mixture. In two other experiments performed under lower blast pressure, disease incidence and severity were significantly lower in mixtures, and yields of the susceptible cultivars grown in mixtures were higher than those of their respective pure stands. Cultivar mixtures are a promising strategy that could contribute to a more sustainable cultivation of rice under upland conditions in the context of subsistence agriculture in Madagascar, where all cropping operations are manual.