This study was focused on meteotsunamis – sea level oscillations with periods similar to those of tsunamis, but caused by meteorological phenomena – and consisted of three parts: first, three numerical models were tested on an idealised basin with synthetic atmospheric forcing and, then, one model was chosen to reproduce several historical meteotsunami events in the Adriatic in two different ways. Models were tested using an idealised rectangular basin of uniform depth. An air pressure disturbance shaped like a sine wave or a boxcar function travelling over the basin was the only external forcing present in the simulations. Three numerical models were used – two finite element models, ADCIRC and SCHISM, and one finite difference model, ROMS. Two parameters were used to quantify results, wave height and energy. Results show that for a sinusoidal disturbance with a wavelength twice the size of the basin or larger, Chrystal resonance develops in the basin causing seiches. The periods of the modelled seiches at the left and right sides of the basin are close to the basin’s eigen period, especially when considering results obtained through maximum wave heights. All model results obtained through maximum energy tend to overestimate the periods. In the central part of the basin where the second mode is expected to be dominant, all the models overestimate the period regardless of the parameter used for calculation. That difference is explained by considering the analytical solution for sea surface elevation obtained from a 1D system of equations for a basin forced by a travelling sinusoidal air pressure disturbance. A sine wave with a wavelength shorter than the basin length causes Proudman resonance to occur. Although all the models give resonant speed close to the speed of free waves in the basin, none of the results is a complete match to the analytical solution. When the wavelength is the same as the basin length or only slightly larger, neither resonance can be seen clearly as a change in regime occurs. These simulations were then repeated for sine waves with the same wavelengths but doubled lengths (that is, two consecutive sine waves travelling over the basin). Results show that these air pressure disturbances transfer more energy into the sea and the seiche periods are closer to theoretical values but Proudman resonance is clearly visible for smaller wavelengths than it was before. While both sets of simulations have similar results, the double sine wave simulations have a more pronounced transition from one resonance to the other as well as a better representation of Chrystal resonance whereas the single sine wave simulations represent Proudman resonance more clearly. Boxcarshaped air pressure disturbances always cause Proudman resonance in the basin regardless of disturbance size,however,not all themodels give resonance at the same speed. The best results are obtained with ROMS which gives resonance at the speed of free waves for every simulation. ADCIRC and SCHISM tend to slightly underestimate the results, especially when considering results obtained through maximum energy. Based on these results, ADCIRC was chosen as the model for further work as it is a finite element model which allows for a very fine resolution inside the bays of interest without using nested domains. SCHISM was rejected because it uses a semi-implicit time scheme which makes choosing an adequate time step slightly more complicated and an inadequate time step can affect Proudman resonance in the model. After the testing phase,modelling off our historical meteotsunami events in the Adriatic was conducted. The simulations used a cut-off depth of 4 m. Results for Vela Luka (21 June 1978), Široka (22 August 2007) and Mali Lošinj (15 August 2008) show that there are significant sea level oscillations in the bays but the amplitudes are, at best, 50% of the observed ones. In the case of Stari Grad (19 February 2010) the modelled oscillations were much smaller, up to 30% of the observed ones. The situation in Stari Grad was very specific that day – sea level was already elevated due to a storm surge and it is assumed that without the storm surge the meteotsunami that hit the bay would not have caused flooding on its own. The next step were simulations that included flooding and drying of grid nodes and they were conducted for the events in Vela Luka and Stari Grad. Grid resolution was enhanced in the entire Adriatic but especially in the two bays where the spatial step was several tens of meters and topography was included in order to allow for wave runup. Results for Vela Luka show a significant improvement from the simulations that used a cut-off depth. The wave amplitude and energy at the mouth of the bay are larger than in cut-off depth simulations, and better resolution inside the bay coupled with flooding and drying caused the modelled wave heights to be twice as large as in previous simulations and closer to the observations. All modelled wave heights were larger than in cut-off depth simulations and this result is an indication that modelling extreme wave heights of destructive meteotsunamis can benefit from a flooding and drying module incorporated into the model. Simulations for the event in Stari Grad were not as successful in reproducing extreme wave heights. Although including the flooding and drying module enlarged the wave height, it was not a significant improvement and there was no flooding in the bay. Since it is assumed that the meteotsunami that hit that day only caused flooding because the sea level was already extremely high due to a storm surge, these results confirm that assumption. This study shows that in order to model meteotsunamis, it is necessary to understand how the model works and how well it reproduces resonant effects connected to such waves. Further improvements could be made with more accurate modelling of atmospheric pressure disturbances, as well as including a flooding and drying module that enables the reproduction of processes that take place once the waves reach the shore.