We thank Steiner for his remarks related to measurement of compliance. We agree that the two measures of compliance used in our study1 (physician's judgment and mean corpuscular volume) are not perfect. This was discussed in the "Comment" section of the article. However, it is important to note that the conclusions of the study were based on the comparisons of compliance between drug users and nondrug users. Even with imperfect instruments, it is unlikely that the compliance rate of drug users was overestimated given the frequent but unsubstantiated belief among physicians that this group has a poor compliance.It was, therefore, reasonable to conclude that the compliance of drug users with zidovudine treatment was not worse than that of nondrug users. But it is also true that better instruments for measuring compliance would be needed to estimate the true magnitude of these compliance rates.