Second language (L2) learners of English whose native languages have relatively simple syllable structure have a strong tendency to modify complex onsets in production. Past studies have shown that such modification is often correlated with sonority-based markedness. According to this principle, the marked bi-consonantal sequences are such that the sonority distance between the first consonant and the subsequent consonant is relatively small. For instance, /pl/ is considered to be less marked than /bl/ since the former has larger sonority distance. A question of interest here is whether such “markedness” would be applicable to the perception of complex onsets by Japanese-speaking learners of English. The current study tested Japanese L2 learners and American English controls in a categorial ABX discrimination test of 8 contrasts between nonsense words with consonant cluster onsets CC(C)VCV vs. CVC(C)VCV sequences (e.g., /spani/ vs. /sepani/) and included /sp, sk, pl, bl, kl, gl, spl, skl/ clusters. Results showed that overall accuracy by Japanese listeners was significantly poorer than for the Americans (72 % and 98% correct, respectively). Certain clusters were harder for Japanese listeners (e.g., 76% correct for /pl/ but 64% for /bl/). However, in general, relative difficulty was not accurately predicted by sonority-based markedness. Alternative hypotheses for relative perceptual difficulties include the acoustic characteristics of the stimulus materials and effects of native phonological structures.