There is a high demand for quantitative information on impacts of climate on crop yields, yield gaps and their variability in Ethiopia, yet, quantitative studies that include an indication of uncertainties in the estimates are rare. A multi-model crop growth simulation approach using the two crop models, i.e. Decision Support System for Agro-Technology (DSSAT) and WOrld FOod STudies (WOFOST) was applied to characterize climate-induced variability and yield gaps of maize. The models were calibrated and evaluated with experimental data from the Central Rift Valley (CRV) in Ethiopia. Subsequently, a simulation experiment was carried out with an early maturing (Melkassa1) and a late maturing (BH540) cultivar using historical weather data (1984-2009) of three locations in the CRV. Yield gaps were computed as differences among simulated water-limited yield, on-farm trial yields and average actual farmers' yields. The simulation experiment revealed that the potential yield (average across three sites and 1984-2009) is 8.2-9.2 and 6.8-7.1 Mg/ha for the late maturing and early maturing cultivars, respectively; ranges indicate mean differences between the two models. The simulated water-limited yield (averaged across three sites and 1984-2009) is 7.2-7.9 Mg/ha for the late maturing and 6.1-6.7 Mg/ha for the early maturing cultivar. The water-limited yield shows high inter-annual variability (CV 36%) and about 60% of this variability in yield is explained by the variation in growing season rainfall. The gap between average farmers yield and simulated water-limited yield ranges from 4.7 to 6.0 Mg/ha. The average farmers' yields were 2.0-2.3 Mg/ha, which is about 1.1-3.1 Mg/ha lower than on-farm trial yields. In relative terms, average farmers' yields are 28-30% of the water-limited yield and 44-65% of on-farm trial yields. Analysis of yield gaps for different number of years to drive average yields indicates that yield gap estimation on the basis of few years may result in misleading conclusions. Approximately ten years of data are required to be able to estimate yield gaps for the Central Rift Valley in a robust manner. Existing yield gaps indicate that there is scope for significantly increasing maize yield in the CRV and other, similar agro-ecological zones in Africa, through improved crop and climate risk management strategies. As crop models differ in detail of describing the complex, dynamic processes of crop growth, water use and soil water balances, the multi-model approach provides information on the uncertainty in simulating crop-climate interactions. (C) 2014 Elsevier B.V. All rights reserved.