Subtidal water levels in Chesapeake Bay, which can have amplitudes as large as 1 m at Baltimore, are an important component of total water levels. The most importance forcing mechanisms for these variations are surface winds over the Bay and coastal subtidal water levels. Two methods for hindcasting subtidal water levels in the Bay were developed: statistical prediction (based on multiple linear regression) and a barotropic numerical circulation model-based prediction. The hindcast water levels were compared with the observed values at three key locations (Chesapeake Bay Bridge Tunnel (CBBT) in the lower bay near the mouth, Solomons Island at midbay, and Baltimore in the upper bay) by a variety of statistical measures. The hindcast results show that in both annually averaged differences and in the incidence of outliers the numerical model-based hindcasts are slightly more accurate than the statistical hindcasts, although on a monthly basis the statistical hindcast was often equal to or better than the model hindcast. Errors in both methods follow a seasonal pattern, being smallest in the summer months and largest in winter. Both hindcast methods explain significant portions of the observed variance. In addition, several techniques were used to estimate the relative importance of coastal water level forcing and wind forcing in the subtidal variability. In general, the two forcings were about of equal importance at Baltimore, while coastal forcing was dominant at Solomons Island and CBBT. ¿ 2001 American Geophysical Union