While early child acquisition – approximately up to age six before the onset of formal education - is a prime example of implicit learning, it is less clear whether and to what extent adult acquisition of a second language (L2) proceeds without intention to learn and in the absence of awareness. In this study, we report on two web-based experiments that investigated to what extent L2 inflectional morphology can be acquired under incidental conditions after a minimal amount of exposure and to what extent the resulting knowledge is implicit or explicit in nature (see Rebuschat, 2013). In both experiments, we used a modification of the experimental design described in Rogers, Revesz and Rebuschat (2015), in which participants were exposed to a semi-artificial language system consisting of foreign words (Croatian nouns) embedded into participants’ L1 (English). While participants were told that they were taking part in a foreign language vocabulary task, the learning target was a case-marking distinction (accusative/dative). Rather than using isolated sentences as the stimulus material during the exposure phase, we embedded the foreign words into a short story (cf. Willems, Frank, Nijhof, Hagoort & van den Bosch, 2015). To determine whether learning outcomes are affected by task demands, we decided to utilize a timed grammaticality judgment task (GJT; Experiment 1) and a timed two-alternative forced choice task (2AFC; Experiment 2). Participants were also administered a Stroop ColourWord test (e.g. Stuss, Floden, Alexander, Levine & Katz 2001) to assess if learning outcomes were related to selective attention and inhibitory control. 200 participants were recruited through CrowdFlower (www.crowdflower.com), an online crowdsourcing platform, allowing us to investigate learning effects in a more diverse population. Performance on the timed GJT (Experiment 1) and the 2AFC task (Experiment 2) served as the measures of learning. Confidence ratings, source attributions, and retrospective verbal reports served to distinguish implicit and explicit knowledge. Interestingly, we found that participant performance was above chance level in the 2AFC task (M = 54.2% (SD=0.5), t(99) = 8.18, p < 0.0001) but not in the GJT (M = 51.4% (SD=0.4), t(99) = 2.18, p > 0.05). The results of mixed logit models revealed that task performance was related to inhibitory control (2AFC: χ2 (1) = 4.87, p < .03; GJT: χ2 (1) = 1.63, p > .05), confidence (2AFC: χ2 (1) = 4.87, p < .03; GJT: χ2(1) = 2.97, p > .05) and source attributions (2AFC: χ2 (1) = 4.87, p < .03; GJT: χ2 (1) = 4.94, p > .05).