Purpose: This study investigated whether past tense use could differentiate children with language impairment (LI) from their typically developing (TD) peers when English is children's second language (L2) and whether L2 children's past tense profiles followed the predictions of Bybee's (2007) usage-based network model. Method: A group of L2 children with LI (L2-LI) and a matched group of L2-TD peers were administered the past tense probe from the Test of Early Grammatical Impairment (Rice & Wexler, 2001) and the Peabody Picture Vocabulary Test (Dunn & Dunn, 1997). A representative input corpus provided distributional information for each verb used. Background information was obtained via parent questionnaire. Results: The L2-LI group used fewer tense-marked verbs than did the L2-TD group. In both groups, vocabulary size and word frequency predicted accuracy with regular and irregular verbs. Children omitted regular past tense marking most often after alveolar stops, dropping the allomorph / Id/; L2-TD children omitted /t/ more often than /d/. Finally, first language typology predicted past tense accuracy. Conclusions: Past tense use could potentially differentiate between English L2 children with and without LI. The impact of vocabulary, frequency, and phonological factors supported the network model and indicated profile differences between L2-LI and L2-TD children.