Long-term and high quality observed climate data are required to analyse with reduced uncertainty temporal and spatial climate variability. The aim of this work is the quality control, the reconstruction and homogenization of monthly precipitation observed in semi-arid basins within Souss-Massa region in North Africa during the period 1933–2010. Data recorded in local meteorological stations underwent a neighbour-based homogenization algorithm. We apply Mann-Kendall test and Sen's method to raw and homogenized time series to calculate trends and their magnitude. The greatest variation in data properties is highlighted for the trend analysis where the homogenization process allows the detection of a higher number of the statistically significant negative trend with 95 and 99% confidence levels.