Singing or listening to songs is one of the most common L2 activities outside the classroom, and many language teachers integrate songs in their teaching; however, the effect of varying song practice modes on L2 learners’ retention of linguistic knowledge is under-researched. Eighty-seven EFL learners of university students were recruited and divided into three subgroups. To avoid different song different effect, the study adopted a counterbalanced designed, in which each subgroup practiced three songs under three modes: listening + singing, listening + reading lyrics orally, and listening only. Three aspects of vocabulary knowledge (aural and written meaning recognition, written form production) were measured in the pre-test, one-week delayed, and eight-week delayed post-tests. HLM was performed to analyze the data. Results showed that the main effects of the four variables (times, practice modes, vocabulary dimensions, songs) had significant effects on L2 learners’ vocabulary retention, but the interaction effects between times and the three other variables were less salient. Overall, students scored higher for singing mode and for written meaning recognition; however, different songs and times seem to have played an important but complicated role on vocabulary retention. Applications of integrating songs in teaching L2 are discussed.