Objective: In the past decade, the U.S. Medicare program has attempted to make its national coverage decisions (NCDs) for new technologies more transparent, evidence-based, and timely. We examined all NCDs from 1999 through 2007 (n=115) to analyze whether decisions were consistent with the evidence and what factors predict review times. Methods: We reviewed NCDs based on publicly-available decision memoranda posted on the Medicare website. We reviewed each NCD on roughly 30 variables, including the quality of clinical evidence available for each technology (i.e., according to sample size, controls, and randomization) and the mean duration of review times. (Medicare does not use cost or cost-effectiveness as a criterion for coverage.) Results: Medicare's 115 NCDs since 1999 have pertained mostly to medical devices (45%), medical/surgical procedures (40%), and pharmaceuticals (9%). The Centers for Medicare and Medicaid Services (CMS) most frequently covered a technology “with conditions” (58%) followed by “no change to existing national coverage” (20%) and “rejected coverage completely” (11%). Only 15% of technologies were supported by what CMS considered “good” quality evidence; in contrast, 41% had “fair” and 36% “poor” evidence. Technologies with good evidence were much more likely to be covered than were technologies with fair or poor evidence (RR=1.80, p=0.0008). NCD reviews averaged 8.7 months (range, 0.5 to 39.2 months). CMS requested input on 16% of NCDs from the Medicare Coverage Advisory Committee (MEDCAC), and on 29% from formal technology assessments (TAs) by evidence-based practice centers. MEDCAC involvement added 5.3 months to review times on average (p=0.0002), and TA involvement added 3.7 months (p<0.0001). Conclusion: Medicare national coverage decisions are generally consistent with the evidence base. The quality of evidence available to CMS for the vast majority of technologies has been fair or poor. Involvement of external advisory bodies is relatively infrequent and tends to prolong review times.