When regression analysis is used as a global spatialisation method for climatic variables, one must pay special attention to the presence of values evading the spatial variation rules stated by the model (outliers). The outliers may alter significantly our regression models, therefore leading us to drawing the wrong conclusions. Our study focuses on the outliers problem through a simple example of mean annual precipitations spatialisation in eastern Romania using the altitude as predictor. The identification of the outliers is based on the magnitude of the residuals, on cross-validation and on the comparison of the regression residuals with the deleted residuals (jackknife error). After the identification stage, we construct regression models leaving out the outliers in order to quantify their negative effects. We then present several possible options to avoid these effects, focusing on the one which eliminates the outliers from the regression models but keeps the residual values in the respective points during the kriging stage in a residual kriging approach.