This paper presents algorithms for extracting the values of relevant parameters from field measurements and 3-dimensional geographical data to be used in neural network modeling of wave propagation loss in microcells. The algorithms extract the feature values from 3-dimensional elevation maps and vector maps based on the theory in Computational Geometry. The neural networks trained on these parameters as their input approximate the function of wave propagation loss and can produce predictions with high accuracy. Some experimental results which show the superior performance of our approach over COST-231 method in actual PCS cell sites operating in the city of Seoul are presented.