Back-interpreting practice refers to interpreting the translated version of a speech back to its source language. The learning opportunities provided by backinterpreting practice can be explained by Swain's (1985) comprehensible output hypothesis, Laufer and Hulstijn's (2001) involvement load hypothesis, and Schmidt's (1990) noticing hypothesis. In the present study, back-interpreting tasks, blended regularly into the undergraduate EFL classroom for 12 weeks, involved 43 high-intermediate to advanced learners working in pairs and interpreting the Chinese translation of English spoken discourse back into English. They then learned useful expressions from the original English discourse before discussing relevant topics in small groups. Two task types--role plays and chain games--were created to avoid monotony. Qualitative data from focus-group interviews were analyzed to understand 12 lower-level speakers' and 11 higher-level speakers' perceptions of the two task types and of the practice in general. Results showed that the format of role plays primed learners to put themselves in the listeners' shoes by focusing on the gist and avoiding literal interpretation. However, the problem of incomprehensibility still existed for some lower-level speakers, and some higher-level speakers might have slacked off by being vague, failing to exhaust their linguistic resources. On the other hand, chain games engaged learners more because higher concentration, faster responses, and language flexibility were required. The format also facilitated more collaboration and peer-learning. Back-interpreting practice exposed learners' blind spots and enhanced their awareness of native-like expressions. The practice may address the challenges of large class size and heterogeneous oral proficiency levels witnessed in EFL contexts.