Climate changes as well as the urbanisation and economic development influence the characteristics of the stormwater runoff in the cities. The sealing of drainage basin surface leads to an increase of the runoff intensity, thereby decreasing the rainwater infiltration. This situation can lead to the risk of flooding in urban areas. Therefore, especially in great cities there is a need for application of such solutions that will support the operation of the sewage systems. The examples of such solutions are, among others, the green roofs. The paper presents the results of investigation of the water retention capacity of 4 green roof models containing following growing media: (1) the typical green roof substrate without any amendments, (2) the substrate with addition of about 1 % by weight of hydrogel (the cross-linked potassium polyacrylate), (3) the substrate containing about 0.25 % by weight of hydrogel, (4) the substrate with addition of expanded clay and perlite. The models were not vegetated in order to investigate only the water retention capacity of drainage elements and substrates. The water retention capacity of green roof models was investigated in the laboratory conditions with use of artificial precipitations simulated after diverse antecedent dry weather periods (ADWP) amounting to: 1, 2, 5, 7, and 12 days. The intensities of artificial precipitations were relatively high and ranged from 1.14 to 1.27 mm/min, whereas their durations ranged from 7.75 to 12.56 min. These values of intensities and durations corresponded to the design rainfall intensities calculated using Blaszczyk’s equation for annual rain depth equal to 600 mm and the return periods ranged from 5 to 15 years. The obtained results indicate that the water retention capacity of green roof models, expressed as the volumes (or depths) of rainwater retained within their structures, increases with an increase of ADWP. Results indicate that the relation between ADWP and the amount of water retained in the layers of green roofs in the case of relatively short antecedent dry weather periods provided for the analysis (from 1 to 7 days) may be approximately linear. The results of the one-way ANOVA indicate that in the case of all models there is a statistically significant difference between the values of retention depth for specified ADWP (p < 0.001). During more than half of simulated precipitations, especially in the case of longer ADWPs lasting 5, 7, and 12 days the best water retention capacity had Model 3, with substrate containing about 0.25 % by weight of hydrogel. On the other hand, the results show that the weakest retention capacity had Model 2 (with substrate containing 1 % by weight of hydrogel). In the case of longer ADWPs (lasting 7 and 12 days) relatively weak water retention capacity had Model 4 (with substrate containing the addition of expanded clay and perlite). It can be concluded that too large amount of hydrogel added to the substrate can have an unfavourable impact on the water retention capacity of green roofs.