Clinical trials endpoints based on magnitude of reduction in HIV-1 RNA levels provide an important complement to endpoints based on either virologic failure or a proportion of patients having HIV-1 RNA levels below a threshold value. However, reductions in HIV-1 RNA often are not completely observed, because many patients have HIV-1 RNA levels below the limit of quantification at the primary follow-up visit. The crude method of analyzing such data is to define all HIV-1 RNA levels that fall below the limit of quantification as being equal to that limit of quantification. This method is widely used even though the underestimation inherent in such a method may also lead to underestimation of treatment difference in terms of HIV-1 RNA reduction. Analyses based on Kaplan-Meier method and censored regression can be used to estimate such a reduction. When a high percentage of patients have HIV-1 RNA levels below the limit of quantification at the time of primary follow-up, which corresponds to censored observations, the Kaplan-Meier method does not always provide an estimate of the median HIV-1 RNA reduction. We discuss a statistical method to provide lower and upper limits of such median reduction or of other percentiles of reduction. We found that when the percentage of censoring is high, the censored method may overestimate the HIV-1 RNA reduction and then may also overestimate the treatment difference. Although the censored method is preferable to the crude method, when the level of censoring is high, we suggest computation of the upper and lower limits either to provide a range of potential values of HIV-1 RNA reduction or to detect overestimation by the censored method.