Buying an apartment is always a big decision, especially when it is the first one and a person wants to become independent. Apartment prices are relatively high, especially in large cities, so buyers want to get to know the real estate market before making a decision. Individual analysis of property prices may be difficult and time-consuming due to the lack of advanced tools, so web portals or real estate agencies use advanced mechanisms to compare them. The traditional methods of price estimation are based on hedonic regression modeling and are nowadays improved by machine learning techniques. The aim of the paper was to present machine learning models, including simple linear regression and decision trees, to predict real estate prices in the 15 largest Polish cities based on their characteristics. The research was done using a dataset containing 27 apartment attributes and their prices. The results showed that both models achieved satisfactory results; R2 equaled 0.72 and 0.79, respectively. Nevertheless, more research is necessary to obtain higher coefficient of determination values. The developed models could be used to predict real estate prices in these cities in an application adapted to Polish market conditions. Further work will include the removal of outliers from features, obtaining and completing the missing data in the feature describing the condition of the apartment, or building separate models for individual cities.