Precipitation amounts of 11 models of the Coordinated Regional climate Downscaling Experiment (CORDEX) are evaluated for northwest Ethiopia against eight sets of gridded observational data sets. The model runs are forced with ERA‐Interim reanalysis and have a resolution of 0.44° while the resolution of the observational data sets varies between 0.0375 and 0.75°. Although the CORDEX ensemble mean overestimates the monthly precipitation compared to observational data sets, observations generally lie within the range of the model ensemble. Models generally overestimate the elevational sensitivity, that is, the simulated rainfall is too large for high elevations and too small for low elevations. The models that use the most smoothened representation of the orography perform the best. A group of six models, responsible for the largest biases, also include poor model orography. Given their strong elevation–precipitation correlation of about 60% even larger biases could be expected in case these models incorporate a correct orography. Also, the sensitivity to the choice of the model domain is investigated. While the eight CORDEX members on the African domain overestimate rainfall, the two members on the South Asia domain underestimate summer precipitation.