The aim of the present manuscript was to investigate the source of congruency effects in weak bilinguals (Experiment 1) and in early language learning (Experiment 2). In both studies, participants performed a bilingual version of a colour-word Stroop task. The standard finding is slower and less accurate responding when the word and colour are incongruent (e.g., “red” in blue) relative to congruent (e.g., “red” in red). This congruency effect occurs for the distracting colour words from both the first and second language. Both stimulus conflict (i.e., conflict between the meaning of the word and ink colour) and response conflict (i.e., conflict between possible response options) contribute to first-language congruency effects. According to some models of early language learning, only one of these two types of conflict should emerge for non-fluent languages. To separate stimulus and response conflict, we used a 2-to-1 keypress assignment manipulation. Interestingly, in one study both stimulus and response conflict were evidenced for the weakly spoken second language (English in native French speakers). In a second study, participants performed a short Croatian colour word learning phase before the Stroop procedure. Stimulus conflict was observed in response times and response conflict in errors for this recently-trained language. These findings suggest that the relatively low-proficient second language words are potent enough to affect semantic identification and response selection.