We examine two explanations of the subliminal affective priming effect. The feelings-as-information model (Schwarz & Clore, 1988) holds that judgements are based on perceptible feelings. Hence, affective influences depend on the source to which feelings are (mis)attributed. In contrast, the affective primary hypothesis (Zajonc, 1980) suggests that affective influences should resist attributional interventions. This is because the affective system responsible for preferences is separate from the cognitive system responsible for inferences; because early affective processes are automatic and therefore inaccessible to higher-order interventions; and because early affective responses are not represented as conscious feelings. We tested these explanations in two experiments that crossed subliminal affective priming with (mis)attribution manipulations. Both studies found reliable shifts in judgements of neutral stimuli as a result of primes even when subjects were aware that their feelings might not be diagnostic for the judgement at hand. Subjects did not report experiencing any feelings in response to the primes. The obtained affective priming effect was independent of response times and subjective reports of engaging in judgemental corrections. However, the priming effect did prove sensitive to the experimental instructions. We discuss the implications of these findings for the affective primacy hypothesis and the feelings-as-information model.