Gridded datasets are of paramount importance to globally derive precipitation quantities for a multitude of scientific and practical applications. However, as most studies do not consider the impacts of temporal and spatial variations of included measurements in the utilized datasets, we conducted a quantitative assessment of the ability of several state of the art gridded precipitation products (CRU, GPCC Full Data Product, GPCC Monitoring Product, ERA-interim, ERA5, MERRA-2, MERRA-2 bias corrected, PERSIANN-CDR) to reproduce monthly precipitation values at climate stations in the Pamir mountains during two 15 year periods (1980-1994, 1998-2012) that are characterized by considerable differences in incorporated observation data. Results regarding the GPCC products illustrated a substantial and significant performance decrease with up to four times higher errors during periods with low observation inputs (1998-2012 with 2 stations on average per 124,000 km2) compared to periods with high quantities of regionally incorporated station data (1980-1994 with 14 stations on average per 124,000 km2). If independent stations were considered, the coefficient of efficiency indicated that only three of the gridded datasets (MERRA-2 bias corrected, GPCC, GPCC MP) performed better than the long term station mean for characterizing surface precipitation. Error patterns and magnitudes show that in complex terrain, evaluation of temporal and spatial variations of included observations is a prerequisite for using gridded precipitation products for scientific applications and to avoid overly optimistic performance assessments.