A stochastic approach to model the spatial variability of primary productivity in a river impoundment is presented in which data from LANDSAT—TM images of reference data from the field were used as basic inputs to the model. Primary productivity is predicted here by a set of variables (chlorophyll, temperature, and turbidity) combined by means of multivated probabilistic model wherein probability distribution functions were assigned to each variable. Linear regression analysis was used to relate the field reference data to the image data used as input to the model. The model was applied using maps of the variables obtained from the digital imagery. The results was a map of primary productivity probabilities converted to absolute values utilizing the cumulative function for field measurements of primary productivity. The field data were collected over °1.5 yr at 16—d intervals at 12 sample sites. The overall r2between the model results using field—measured variables and the measured primary productivity carbon values was 0.85 with an RMS (root mean square) error of 16 mg°m—3°h—1, where the carbon values ranged from 4 to 300 mg.m—3.h—1. To test the methodology for the generation of primary productivity maps from LANDSAT imagery, two maps of primary productivity for Kentucky Lake were generated from data collected in August and December 1988. The model output was, in turn, compared to primary production carbon measurements from the reservoir. The r2s were 0.89 and 0.76 with RMS errors in these carbon values of 10 and 18 mg°m—3°h—1. No specific assumptions about Kentucky Lake were required for this approach, so the methodology is applicable to other lakes or reservoirs in similar trophic states.