Just as happy people see the proverbial glass as half-full, optimistic or pessimistic responses to ambiguity might also reflect affective states in animals. Judgement bias tests, designed to measure these responses, are an increasingly popular way of assessing animal affect and there is now a substantial, but heterogeneous, literature on their use across different species, affect manipulations, and study designs. By conducting a systematic review and meta-analysis of 459 effect sizes from 71 studies of non-pharmacological affect manipulations on 22 non-human species, we show that animals in relatively better conditions, assumed to generate more positive affect, show more optimistic judgements of ambiguity than those in relatively worse conditions. Overall effects are small when considering responses to all cues, but become more pronounced when non-ambiguous training cues are excluded from analyses or when focusing only on the most divergent responses between treatment groups. Task type (go/no-go; go/go active choice), training cue reinforcement (reward-punishment; reward-null; reward-reward) and sex of animals emerge as potential moderators of effect sizes in judgement bias tests.