The performance of the spectral wind wave model SWAN in tidal inlet seas was assessed on the basis of extensive wave measurements conducted in the Amelander Zeegat tidal inlet and the Dutch Eastern Wadden Sea, as well as relevant data from other inlets, lakes, estuaries and beaches. We found that the 2006 default SWAN model (version 40.51), the starting point of the investigation, performed reasonably well for measured storm conditions, but three aspects required further attention. First, over the near‐horizontal tidal flats, the computed ratio of integral wave height over water depth showed an apparent upper limit using the default depth‐limited wave breaking formulation and breaker parameter, resulting in an underprediction of wave heights. This problem has been largely solved using a new breaker formulation. The second aspect concerns wave‐current interaction, specifically the wave age effect on waves generated in ambient current, and a proposed enhanced dissipation in negative current gradients. Third, the variance density of lower‐frequency wind waves from the North Sea penetrating through the inlets into the Wadden Sea was underpredicted. This was improved by reducing the bottom friction dissipation relative to that of the default model. After a combined calibration, these improvements have resulted in a relative bias reduction in Hm0 from −3% to −1%, in Tm−1,0 from −7% to −3%, and in Tm01 from −6% to −2%, and consistent reductions in scatter, compared to the 2006 default model. Key Points An extensive data set for wave model evaluation was collected in the Wadden Sea The wave model SWAN generally performs well in tidal inlet seas, with exceptions Depth‐induced breaking, wave‐current interaction and propagation were improved