Threat conditioning research frequently relies on making inferences about associative learning mechanisms using a single item for each conditioned stimulus (CS+, CS-) and the unconditioned stimulus. However, this type of design does not allow for assessing the extent to which results generalise to populations of items and can lead to anticonservative inferences. In addition, employing a small number of trials when measuring noisy, psychophysiological responses may be problematic for establishing adequate signal-to-noise ratios. This study addresses these methodological issues and provides means for measuring associative learning and extinction mechanisms in a generalisable manner, using pupillometry as well as ratings of contingency awareness, valence, and arousal. To maximise the number of items and trials per condition, we employed an auditory blocked conditioning paradigm in which learning and extinction were established several times using different sets of conditioned (pure tones) and unconditioned stimuli (environmental sounds). We used cluster-based permutation to identify pupillary time windows of interest and linear-mixed effects modelling to account for by-subject and by-item variability. We demonstrate highly correlated early and late pupillary conditioned responses during acquisition, suggesting a common underlying learning mechanism. These responses habituated immediately during extinction. We also demonstrate (rating tasks) that conditioned valence responses are less susceptible to extinction and potentially index an additional learning mechanism. By contrast, subjective arousal ratings were less sensitive to associative learning. Finally, we demonstrate the potential risk of alpha inflation when using ‘traditional’ ANOVA-based analyses on our data.