Yield gaps, and factors causing them, are of prime importance for agricultural production. There are too many yield-constraining factors in the variable rainfed lowland rice agroecosystems to be studied in full factorial experiments. An empirical-exploratory attempt is presented to characterize yield gaps and attainable yield levels as related to agroecosystem variables and to adjust yield expectations and identify research and management needs. Data on crop and pest management practices, soil conditions, weather, crop performance, and biotic and abiotic stresses were collected in over 600 plots in farmers’ rainfed lowland ricefields in northwest Luzon (Philippines), northeast Thailand, and the Mekeong River delta (south Vietnam) from 1992 to 1994. The CERES-Rice simulation model was used to estimate weather and nitrogen (N) limited attainable yield levels, while a simple empirical model was used to estimate yield trends based on fertilizer N and soil organic matter. Results of simulation runs indicated that the weather-adjusted yield gaps, i.e. the deviations of the observed yields from the weather-limited simulated yields, averaged about 35% in the Philippines, 45% in Vietnam, and 55% in Thailand. They were mainly due to N limitation in Thailand, where soil-N and fertilizer use is low, while in the other two countries they were mainly related to other constraints. In multiple regression analyses, terms related to soil carbon content and/or amount of N fertilizer captured the main yield trend well. The remaining yield variation could partly be explained by interactions between these main terms and severity levels of diseases and pest damage, water stress, and other variables related to soil conditions, crop and pest management, and weather. Factor and canonical correspondence analyses gave further insights into links among actual yield, estimated attainable yield levels and corresponding yield gaps, and other agroecosystem variables. The approach is useful for quantifying attainable yield levels and yield gaps at various constraint levels and leads to a better understanding of complex relationships between agroecosystem variables, improved yield expectations, and a better understanding of the role of yield-determining factors. This can be instrumental for prioritizing research on yieldlimiting factors and guiding crop and pest management decisions.