Although in recent years some studies have found evidence suggesting that working memory (WM) may operate on unconscious perceptual contents, decisive demonstrations of the existence of unconscious WM are lacking. In the present Registered Report, we replicate the first study on this topic by Soto et al. (Working memory without consciousness. Curr Biol 2011;21:R912–3.): a visual discrimination task asking participants to report the direction in which a subliminal Gabor grating was rotated after a 2-s delay. We acquired a multisite sample from 19 laboratories, with a larger number of participants (N = 531) and trials (720 in two sessions) than those typically used in previous studies. As a result, a large-sample, international, and open-access dataset is now available for researchers and future analyses. Furthermore, some minimal baseline requirements were guaranteed for the experimental task (i.e. number of valid trials, motivation, and consistent labels for the Perceptual Awareness Scale). The results showed (1) above-chance WM performance in cue-present trials reported as unseen (.55 accuracy), (2) a significant positive correlation between WM performance and cue detection sensitivity (r = .228), and (3) a significant above-chance intercept in the regression of performance on sensitivity (β0 = .521). These findings suggest that WM can operate on unconscious representations, although it remains positively associated with perceptual sensitivity. Crucially, because measurement error could compromise the interpretation of these three results, we provide evidence for our measures’ excellent reliability and, more fundamentally, for their validity.