A numerical simulation of the ground-water flow system in the Castle Lake debris dam, calibrated to data from the 1991 and 1992 water years, was used to estimate factors of safety against heave and internal erosion. The Castle Lake debris dam, 5 miles northwest of the summit of Mount St. Helens, impounds 19,000 acre-ft of water that could pose a flood hazard in the event of a lake breakout. A new topographic map of the Castle Lake area prior to the 1980 eruption of Mount St. Helens was prepared and used to calculate the thickness of the debris avalanche deposits that compose the dam. Water levels in 22 piezometers and discharges from seeps on the dam face measured several times per year beginning in 1990 supplemented measurements in 11 piezometers and less frequent seep discharge measurements made since 1983. Observations in one group of piezometers reveal heads above the land surface and head gradients favoring upward flow that correspond to factors of safety only slightly greater than 2. The steady-state ground-water flow system in the debris dam was simulated using a threedimensional finite difference computer program. A uniform, isotropic model having the same shape as the dam and a hydraulic conductivity of 1.55 ft/day simulates the correct water level at half the observation points, but is in error by 10 ft or more at other points. Spatial variations of hydraulic conductivity were required to calibrate the model. The model analysis suggests that ground water flows in both directions between the debris dam and Castle Lake. Factors of safety against heave and internal erosion were calculated where the model simulated upward flow of ground water. A critical gradient analysis yields factors of safety as low as 2 near the piezometers where water level observations indicate low factors of safety. Low safety factors are also computed near Castle Creek where slumping was caused by a storm in January, 1990. If hydraulic property contrasts are present in areas of the debris dam unsampled by piezometers, then low safety factors may exist that are not evident in the numerical model analysis. Numerical model simulations showed that lowering Castle Lake by 40 feet increases many factors of safety by 0.1, but increases greater than 1 are limited to the area of 1990 slumping.