Research suggests that second language words are learned faster when they are similar in phonological structure or accent to the words of an individual's first language. Many major theories suggest this happens because of differences in frequency of exposure and context between first and second language words. Here, we examine the independent contribution of accent and phonological structure on the speed of word learning and on the depth of semantic encoding while controlling for frequency of exposure and context. Fifteen participants learned novel words associated with abstract shapes in a paired associates task. The words systematically varied in their accent and phonological structure. Learning speed was measured for each word and the depth of semantic encoding was measured via a novel manipulation of the N300 ERP component in a Picture Recognition Task of the learned items. Both non-native structure and accent slowed word learning and differences in the N300 effect indicated that semantic encoding was shallower for words with a non-native than native phonological structure, despite almost ceiling levels of accuracy. These results are consistent with a model of second language learning that proposes both accent and phonological structure influence how fast and deep new language vocabulary is learnt.