In recent years, there has been a common use of automatic precipitation mapping based on various interpolation methods. Based on multiple linear regression, the study primarily focuses on the entire Cheliff basin (43 750 km²), to search a suitable model for mapping mean annual precipitation based on 89 rainfall stations of the observed series (1968/69–2001/2002). Then conducts on the upper part of Cheliff basin downstream the Boughzoul dam (4777 km²) using 11 rainfall stations of annual precipitation series relative to the period 1982/83–2020/2021. The same approach was used to map median annual precipitations and quantiles of rain associated with the return periods of 5 and 10 years, representing both wet and dry periods. Indeed, this work is based on the backward elimination method between the observed annual precipitation and four predictors: smoothed altitude (Zs), longitude (X), latitude (Y) and distance from the sea (D). The model’s results are satisfactory with a global correlation coefficient for both regional (R = 0,89) and local scale (R > 0,81). The final maps obtained are produced after interpolating by kriging the residuals. The median annual precipitation map was compared to that elaborated by the National Hydraulic Resources Agency (ANRH). The comparison results of 2296 nodes of the grid covering the Upper Cheliff basin showed negative deviations on average - 10% and reaching - 34% in northern area reflecting a dry trend in precipitation. The final grids of rainfall can be integrated into Geographical Information Systems related to many sectors including water management and climate change.