This study investigated the relationship between lexical features of listening texts and listening comprehension with a mix of quantitative and qualitative methods. Totally 73 listening test papers completed by S5 students (Grade 11) from a secondary school in Hongkong were collected for data analysis. A vocabulary test provided a baseline about the students’ proficiency. Function words and content words in the text were operationalised by Sixltr and Analytic, two variables computed by LIWC, an auto text analysis tool. Rasch item person map was generated by using an auto test analysis tool. The results of vocabulary test scores, item analysis and content analysis of the listening test paper and student output were discussed with reference to the first lexical threshold at a 2000-word level. Regression analysis suggested that three lexical variables - lexical frequency, Sixltr (content words) and Analytic (function words) - explained 29.7% of the variance in the listening comprehension. Vocabulary test explained 66.6% of the variance in the listening scores. Content analysis with item person maps provided insights into students’ stage of language development. The findings led to a listening threshold hypothesis. Listening processing became increasingly automatic when students approached the lexical threshold. They were more sensitive to redundant and grammatical details in the clues. To act on the hypothesis, a plan of achieving threshold with simpler texts, focused practices and listening tasks has been proposed. Five types of focused practices have been recommended to improve students’ word recognition and listening fluency.