Aim: Phe aim of the article is to present CrowdSim, proprietary software for modeling the dynamics of pedestrians. The software offers algorithms based on a 2-dimensional model of a Cellular Automaton (CA). They include strategic pedestrian choices of the exit (CA-A model) and the formation of blockages (Model CA-B). Implementation of various models in terms of their complexity provided the opportunity to compare their efficiency, functioning and realism of the results obtained. Project and methods: In order to better and more credibly determine pedestrian paths, two modifications were proposed to improve the operation of simulation algorithms. The first of these was the introduction of advanced algorithms ranking cells: SRA (Simple Ranking Algorithm), BFS-RA (Breadth First Search Ranking Algorithm) and ARA (Advanced Ranking Algorithm). The value of the ranking of a given cell in the simulation process is for each of the actors the determinant of the path on which it moves. Another proprietary improvement was the introduction of the concept of actor's memory to the definition of a cellular automata, thanks to which pedestrians, while remembering their last movements, move more smoothly and realistically. Results: For the purposes of model comparison, a special map of the building with the dimensions of 40 to 22.5 meters (900 m2) has been prepared, able to accommodate a maximum of 3000 people. The simulations showed that in the case of the CA model, the actors gather at one of the outputs while the others are mostly unused. This is due to the fact that the CA model lacks the cost function, so the actor always chooses the nearest output. Things are different in the case of the CA-A model. Here, the actors equally burden each exit, they are able to make decisions regarding the choice of the exit not only on the basis of the distance but also the concentration of actors at the given exit. When comparing the CA-A and CA-B models, there are practically no differences in the received output load statistics. Only after a closer look at the movement of the actors during the simulation can be seen that if in the case of a clean CA-A model people move according to the ideal order, in the CA-B model there are momentary blockages and deadlocks of actors. Conclusions: CrowdSim software created and presented in this work can be used by designers, architects or constructors as well as law enforcement, protection of a given facility or organizers of mass events. Simulations made repeatedly for one scenario with different input parameters reveal problems and bottlenecks that may appear during evacuation from an object at risk. Conducting such simulations at the design stage can have a very positive impact on the safety of people using a given facility, as well as improve its functionality.