Abstract This article examines the issue of cognates in frequency-based vocabulary size testing. Data from a pilot study for a cognate-controlled English vocabulary size test was used to assess whether a group of Japanese university English learners (n = 60) were more successful at responding to cognate items than noncognate ones in three 1000 word frequency bands on a Japanese-English translation task. The results showed a statistically significant difference between scores achieved on cognate and noncognate items at the 2000 and 3000 frequency levels, but not at the 1000 frequency level. The findings suggest that cognate items may be easier for test-takers to respond to than noncognate ones of similar frequency, indicating the importance of ensuring that their respective proportions in tests are representative of those inherent in the frequency bands they have been sampled from. It is also argued that such representativeness may best be achieved via a stratified item sampling approach.