The presented research concerns methods related to reduction of elevation data contained in digital terrain model (DTM) from airborne laser scanning (ALS) in hydraulic modelling. The reduction is necessary in the preparation of large datasets of geospatial data describing terrain re lief. Its course should not be associated with regular data filtering, which o ften occurs in practice. Such a method leads to a number of important forms important for hydraulic modeling being missed. One of the proposed solutions for the reduction of elevation data contained in DTM is to change the regular grid into the hybrid structure with regularly distributed points and irregularly located critical points. The purpose of this paper is to compare algorithms for extracting these key points from DTM. They are used in hybrid mod el generation as a part of elevation data reduction process that retains DTM accuracy and reduces the size of output files. In experiments, the following algorithms were tested: Topographic Position Index (TPI), Very Important Points (VIP) and Z - tolerance. Their effectiveness in reduction (maintaining the accuracy and reducing datasets) was evaluated in respect to input DTM from ALS. The best results were obtained for the Z - tolerance algorithm, but they do not diminish the capabilities of the other two algorithms: VIP and TPI which can generalize DTM quite well. The results confirm the possibility of obtaining a high degree of reduction reaching only a few percent of the input data with a relatively low decrease of vertical DTM accuracy to a few centimetres.