This paper is an extension of the dynamic duality theory established by Epstein (1981) and McLaren and Cooper (1980). Applications of the duality theory to dynamic optimization problems provide an efficient means of constructing empirically implementable models. Nevertheless, implications drawn from the dynamic dual models are subject to Lucas’ criticism due to the commonly maintained assumption of static price expectations. Explicitly incorporating the least-squares learning mechanism popularized by Marcet and Sargent (1989a), this paper establishes the duality between the firm’s production technology and the value function under nonstatic price expecations. Complete characterization of the value function is derived to infer dynamic decision rules consistent with the rationale of learning.