The worth of additional data to a digital model of the Tucson basin, Arizona, was computed by using a basic form of statistical decision theory. The model variables for which additional data were hypothesized included the aquifer coefficients of storage and transmissivity, initial water levels, discharge, and recharge. The worth of data was evaluated in terms of the expected reduction in error in predicted water levels associated with collection of more data on one variable at one point in the model. A limited number of tests suggested that the Tucson basin model could be improved most by obtaining more data on discharge and recharge in areas where these variables were large and by obtaining more data on transmissivity where it was uncertain. More data on initial water levels and storage coefficient commonly were less helpful. The sensitivity of the results to the assumptions made in postulating discrete frequency distributions with largely subjectively determined parameters for the model variables was estimated. These tests indicated that the numerical values for worth of data were sensitive to the assumptions but that the relative rankings of variables in terms of worth of added data remained constant.