The reanalysis temperature products are commonly used for studying climate changes from short to long-term scales at regional and global levels, but yet not in most developing countries. This study was conducted in the Horn of Africa region, particularly in Ethiopia, from 1990-2020. To evaluate the performances of the two global reanalysis temperature products, the study compared temperature products with 37 stations of ground-based observed data. This study used five statis-tical error metrics, namely Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Percent Bias (PBIAS), Correlation Coefficient (r), and Kling Gupta Efficiency (KGE). Moreover, scatter plots and line graphs were used to evaluate the model performances. For filling the missing values of daily temperature data, Multivariate Imputation by Chained Equations (MICE) technique was used, and a point-to-pixel evaluation approach was employed. The study indicated MERRA v2.0 has better performances for simulating maximum and minimum temperatures at monthly, seasonal, and annual temporal scales, and ERA5 shows better performances at daily time steps over most agro-ecological zones by the majority of evaluation scores. Moreover, MERRA v2.0 exhibited the lowest PBIAS values at monthly, seasonal (Kiremt), and annual time scales over most agro-ecological zones. In general, MERRA v2.0 performs better than ERA5 over most agro-ecological zones by most evaluation metrics. Therefore, this study recommends MERRA v2.0 reanalysis dataset for estimating temperature in areas where observed meteorological data are scarce for monthly, seasonal and annual scale of the country.