ITT (analysis by protocol-assigned rather than actual therapy) is the gold standard for clinical trials. Yet, ITT has been criticized as overly conservative and a potential cause of false-negative conclusions for major trials like the NNST involving highly controversial therapies and virtually unavoidable treatment protocol violations. In analyzing NNST, we compared standard ITT to an analysis designed to adjust for treatment noncompliance. NNST, a 12-center, 34-month, double-masked trial, compared early (14-28d) with late (28-42d, if needed) steroids (S) for infants ≤1500 g BW and ventilator dependent at 14d. Survivors were given an extubation score (ES) equal to the days from randomization to successful extubation. (Deaths received a score higher than that of any survivor.) By ITT, ES did not differ significantly between the early and late groups (medians 36.0 vs. 36.5; Wilcoxon p=0.18). Yet, 13% of the late group received open-label S between 14 and 28 d, a protocol violation that might bias ITT toward the null hypothesis. In the adjusted analysis, the ES of the 13% were replaced by their expected values had they remained under protocol. These values were estimated from a model based on late group infants who did not receive open-label S, using pre-randomization variables selected by bootstrap. This yielded a median ES for early and late groups of 36.0 and 36.0 d, respectively (Wilcoxon p=0.22). Using a similar approach for death or ventilator therapy at 120 d, the p was 0.87 by the adjusted analysis and 0.25 for ITT. For NNST, a trial involving great effort to maximize compliance as well as an increase in sample size in anticipation of 10% noncompliance, ITT is supported by the adjusted analyses. The value of such analyses deserves exploration in other trials.