Regional-scale hydrology of the fresh water lens in the Northern Guam Lens Aquifer has been simulated in the past using a finite element, sharp interface computer model, SWIG2D. Systematic differences exist between observed and computed water levels. Computed seasonal peak water levels are higher, and the computed seasonal lows are lower than the respective observed levels. It is hypothesized that vadose storage must store a substantial amount of water during the wet season and release it gradually into the lens during the dry season. Flow through the vadose zone was simulated with a one-dimensional finite element, unsaturated flow program, UNSAT1D, in which the van Genuchten model is used to characterize unsaturated diffuse flow through the matrix of the vadose zone. An additional parameter (SINK) was added to the van Genuchten set to account for rapid infiltration down open pathways (fractures) associated with the closed depressions of the karst terrain. A global-optimization technique (Shuffled Complex Evolution, or SCE-UA, Method) was used to obtain the parameters that minimized the difference between simulated and observed water levels. Simulations incorporating the van Genuchten model were accomplished by combining the two programs, UNSAT1D and SWIG2D, into a single program. The sum-of-squared-errors (SSE) between computed and observed water levels in four observation wells was minimized using SCE-UA, reducing the arithmetically-averaged SSE of the four wells by 30% compared with the SSE obtained when the vadose zone was not modeled. These results suggest that vadose storage is significant. On the other hand, the fact that the best fit obtained with an optimum parameter set was able to reduce the SSE by no more than 30% suggests that additional phenomena have yet to be accounted for to more fully explain differences between simulated and observed well water levels.