Given the high cost of data collection at groundwater contamination remediation sites, it is becoming increasingly important to make data collection as cost-effective as possible. A Bayesian data worth framework is developed in an attempt to carry out this task for remediation programs in which a groundwater contaminant plume must be located and then hydraulically contained. The framework is applied to a hypothetical contamination problem where uncertainty in plume location and extent are caused by uncertainty in source location, source loading time, and aquifer heterogeneity. The goal is to find the optimum number and the best locations for a sequence of observation wells that minimize the expected cost of remediation plus sampling. Simplifying assumptions include steady state heads, advective transport, simple retardation, and remediation costs as a linear function of discharge rate. In the case here, an average of six observation wells was needed. Results indicate that this optimum number was particularly sensitive to the mean hydraulic conductivity. The optimum number was also sensitive to the variance of the hydraulic conductivity, annual discount rate, operating cost, and sample unit cost. It was relatively insensitive to the correlation length of hydraulic conductivity. For the case here, points of greatest uncertainty in plume presence were on average poor candidates for sample locations, and randomly located samples were not cost-effective. ¿ American Geophysical Union 1994