A multivariate regression method to reconstruct annual and monthly global precipitation anomalies for the years 1850-2012 based on station observations of the Global Historical Climatology Network (GHCN) and satellite data of the Global Precipitation Climatology Project (GPCP) was applied and validated. The GPCP dataset is used to derive Empirical Orthogonal Functions (EOFs) which are used as regressors in the reconstruction. As a model selection criterion, Akaikes Information Criterion is found to be not applicable, as it suggests to include every mode, and therefore noise as well. Hence, a cumulative variance criterion is used instead. The application shows the method's ability to map and reconstruct ENSO-related patterns on annual and seasonal basis. A time series of annual average precipitation anomalies shows a trend of 0.013(mm/day)/100a starting in 1900. The thesis shows how reducing the number of included modes is a useful tool when dealing with sparsely covered observation fields. In the validation, a temporal dependence in the regression residuals is found. The residual variance is negatively correlated to the number of stations. Residual normality holds for 97% of the years after removing outliers.