L1-L2 translation of separate sentences is one kind of task format used by mainstream EFL teachers to assess their learners' grammatical accuracy. Aimed at improving teacher-written translation items, this study analyses linguistic features potentially causing such decontextualized cues (and their target responses) to sound odd or untypical of naturally used language. The findings show that some items elicit structures that are marked in their contexts, others are translationally overchallenging, and yet others have textual flaws. The following research method was employed: English-Hebrew bilinguals, mostly EFL teachers, translated seven cues from Hebrew to English. This resulted in numerous translation options for each cue. The analysis of these multiple translations and their comparison with corpus frequency data yielded most of the above-mentioned findings. In addition, another group of Hebrew-English bilinguals translated the previous group's most frequent translations back into Hebrew. This added information about the textual quality of the Hebrew cue.