There are two main memory systems: declarative and procedural memory. Knowledge of these two systems in fish is scarce, and controlled laboratory studies are needed. Trace classical conditioning is an experimentally tractable model of declarative memory. We tested whether rainbow trout (Oncorhynchus mykiss) can learn by trace conditioning and form stimulus-stimulus, as opposed to stimulus-response, associations. We predicted that rainbow trout trained by trace conditioning would show appetitive behaviour (conditioned response; CR) towards the conditioned stimulus (CS; light), and that the CR would be sensitive to devaluation of the unconditioned stimulus (US; food). The learning group (L, N = 14) was trained on a CS + US contingency schedule with a trace interval of 3.4 s. The control group (CtrL, N = 4) was kept on a completely random schedule. The fish that learnt were further trained as either an experimental (L, N = 6) or a memory control (CtrM, N = 3) group. The L group had the US devalued. The CtrM group received only food. No fish in the CtrL group, but nine fish from the L group conditioned to the light. When tested, five L fish changed their CRs after US devaluation, indicating learning by stimulus-stimulus association of the light with the food. CtrM fish retained their original CRs. To the best of our knowledge, this experiment is the first to show that rainbow trout can learn by trace classical conditioning. The results indicate that the fish learnt by 'facts-learning' rather than by reflex acquisition in this study.