The validity of applying the signal detection theory (SDT) in measuring native Japanese listeners' sensitivity to large sets of English /r/ and /l/ tokens was tested in a one-interval identification task with confidenceratings. Two distinct sets of 200 tokens (100 /r/'s and 100 /l/'s produced by ten native English talkers) were used for each of the positions in which the target consonants appeared (i.e., word initial and consonant cluster). The response patterns of all 20 subjects (10 for word initial, 10 for cluster positions) except one in the word initial condition were consistent with the prediction of SDT, indicating that the application is valid for the majority of Japanese listeners. The two distinct sets of stimuli for each position condition, however, did not lead to identical results, reflecting strong observer--stimulus interaction. Reliability of this sensitivity estimation was also confirmed by comparing two blocks of presentations (200 trials each) for each stimulus set. The present paradigm thus offers a valid and reliable method of estimating native Japanese listeners' sensitivity to a given large set of English /r/ and /l/ tokens. The underlying mathematical model of SDT and the interpretation of its statistics such as d[inf a] (sensitivity) and c[inf a] (response bias) are shown.