High altitude upland rice production systems are expected to benefit from climate change induced increase in temperatures. The potential yield of rice genotypes is governed by the thermal environment experienced during crop development phases when yield components are determined. Thus, knowledge on genotypic variability in phenotypic responses to variable temperature is required for assessing the adaptability of rice production to changing climate. Although, several crop models are available for this task, genotypic thermal constants used to simulate crop phenology vary strongly among the models and are under debate. Therefore, we conducted field trials with ten contrasting upland rice genotypes on three locations along an altitudinal gradient with five monthly staggered sowing dates for two years in Madagascar with the aim to study phenological responses at different temperature regimes. We found that, crop duration is equally influenced by genotype selection, sowing date and year in the high altitude. In contrast, in mid altitudes genotype has no effect on crop duration but year and sowing date strongly affect crop duration. At low altitudes crop duration is more affected by sowing date and less by genotype and year. Every 1°C increment in mean air temperature decreases crop duration (germination to flowering) by 5 to 9 days depending on genotype. Using a wide range of environments for estimating thermal constants (Tbase and Tsum) allowed for more accurate results under field conditions. Whereas the mid altitudes represent favorable conditions for upland rice, grain yield is strongly affected by low temperatures at high altitudes and severly influenced by frequent tropical cyclones at low altitudes. In high altitude, genotype explained 68% of variation in spikelet sterility, whereas in mid and low altitudes environment explained more than 70% of the variation. The phenological responses determining crop duration and yield, the reported basic genotypic thermal constants, and the analyses of genotypic thermal responses with regard to spikelet sterility reported here, provide valuable information for the improvement of rice phenological and growth models urgently needed to develop new genotypes and better adapted cropping calendars.