The purpose of this study was to learn how transparency and ambiguity affect idiom learning. To start, 157 French idioms were translated to English and normed for familiarity, transparency, and ambiguity. Experiment 1 was a training study in which 32 of these idioms were taught to 25 native English speakers over two days of training. A cued recall test during a third session showed a reliable effect of transparency, but performance was close to ceiling. In Experiment 2, the amount of training was reduced to one session and a semantic relatedness test was included after the cued recall test. The results of Experiment 2 suggest that high transparency idioms are recalled with greater accuracy in a cued-recall test but low transparency idioms are recalled with greater accuracy in a semantic relatedness test. No significant effect of or interaction with ambiguity was found.