Cognitive bias tasks purport to assess affective states via responses to ambiguous stimuli. We hypothesized that a novel cognitive bias task based on positive reinforcement using quantity differences would detect changes in affect in captive grizzly bears (Ursus arctos horribilis). We trained bears (n = 8) to respond differently (nose or paw touch) to two stimuli (light or dark gray cue cards), with responses counterbalanced across bears. The two cues signaled a small or large food reward, respectively. Responses to ambiguous probe stimuli (i.e., shades of gray) intermediate to the trained stimuli were classified as either 'optimistic,' appropriate for the larger reward, or 'pessimistic,' appropriate for the smaller reward. In Experiment 1, we explored the contrast in reward size necessary to detect a change in response across probe stimuli (large reward, 3 or 6 apple slices: small reward, 1 slice). We observed a change in response across probe stimuli, with no difference in response between reward-value groups, indicating that a contrast of 3:1 apple slices was sufficient to affect responses. In Experiment 2, we investigated cognitive bias after 2.1 h of exposure to enrichment items varying in attractiveness. Results were unaffected by enrichment type or time spent interacting with enrichments, indicating that the task failed to demonstrate criterion validity for comparing mood following exposure to different enrichment items. However, greater time spent pacing prior to testing was associated with 'optimistic' judgments. The data provide some support for use of cognitive bias tasks based on quantity differences in animal welfare assessments involving captive wildlife.