Satellite rainfall products are an important source of rainfall data in un-gauged catchments. However, these products need to be validated as their accuracy can be affected by geographical position, topography, climate and embedded algorithms. Eight satellite rainfall products such as African Rainfall Climatology (ARC2), Climate Hazards Group InfraRed Precipitation with Stations (CHIRPs), Global Precipitation Climatology Project GPCP), CPC Morphic technique (CMorph), Atmospheric Administration Climate Prediction Center (NOAA-CPC) merged analysis (CMap), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), African Rainfall Estimation Algorithm version 2 (RFEv2) and Tropical Rainfall Measurement Mission (TRMM) were evaluated against ground observations over the complex topography of the upper Tekeze-Atbara basin in Ethiopia. The accuracy of the datasets was evaluated at different temporal and larger spatial scales over the period 2002–2015. The results show that the rainfall data of CHIRPS outperformed all other products at all temporal and spatial scales. Next to CHIRPS, estimates from RFEv2, 3B42v7, and PERSIANN products are closest to the measurements at rain gauges for all spatiotemporal scales: daily, monthly and seasonal, and both at point and spatial scales. The percentage bias (PBias) and correlation coefficient (r) of these products were within ±25% and >0.5 for all scales. The remaining products performed poorly with PBias up to 200% and lower r (<0.5) at all scales. However, the performance of all products improved as the temporal scale increased to month and season at all spatial scales. Compared to low altitudes <2000 m above sea level (m.a.s.l.), the PBias at high altitude (>3000 m.a.s.l.) increased by 35% whilst r dropped by 28%. CHIRPS and 3B42v7 products showed best agreement in mountainous terrains. However, all datasets show no consistency of the error sign. CMorph and 3B42v7 consistently overestimate rainfall relative to all rain gauges during the pixel-to-point rainfall comparison approach and at lowland areas during the areal averaged rainfall comparison. The other six products showed a clear underestimation at all spatial scales. In summary, the results show that rainfall estimates by CHIRPS, RFEv2 and 3B42v7 have a consistently better agreement with ground rainfall at all spatiotemporal scales. Considering the complex topography and limited gauges, the performance of CHIRPS, RFEv2 and 3B42v7 indicates that these products can be used for hydrological and overall water management applications in the region.