Cognitive bias tests are frequently used to assess affective state in nonhumans. We adapted the ambiguous-cue paradigm to assess affective states and to compare learning of reward associations in two distantly related species, an American black bear and three Western lowland gorillas. Subjects were presented with three training stimuli: one that was always rewarded (P), one that was never rewarded (N) and one that was ambiguous (A) because its reward association depended on whether it had been paired with P (PA pairing) or N (NA pairing). Differential learning of NA and PA pairs provided insight into affective state as the bear and one gorilla learned NA pairs more readily, indicating that they focused on cues of reinforcement more than cues of non-reinforcement, whereas the opposite was true of one gorilla. A third gorilla did not learn either pairings at above chance levels. Although all subjects experienced difficulty learning the pairings, we were able to assess responses to A during probe trials in the bear and one gorilla. Both responded optimistically, but it was difficult to determine whether their responses were a true reflection of affective state or were due to preferences for specific stimuli.