Principles in foraging and standard associative learning theories motivate a model for Pavlovian conditioning. The model tracks distal and proximal scales of expected reward probabilities plus the strength of signal-reward association. A combined reward probability is developed by combining the distal and proximal estimates through their uncertainties. Possible neural structure equivalents to the model variables are discussed. Model flexibility is demonstrated with data on the partial reinforcement extinction effect, a phenomenon difficult to explain with learning models.