The key objective of the paper was to describe quantitative relationships between the concentrations of different forms of aluminum and the values of the remaining parameters measured in soils of Beskid Mountains. The analyzed environmental data contain outlying objects, i.e. parameters' values resulting from measurements errors, faults in data acquisition or drastic changes in the environment. As classical calibration methods are very sensitive to the presence of outliers in data sets, they result in the construction of improper models. In this paper the PLS-EP models, based on the evolution program, were constructed to overcome the problem of outlying objects in environmental data sets.