We applied a wide array of statistical models to re-examine the conclusions from 23 meta-analyses of diagnostic tests selected randomly from all identified from a Medline search through 2003. After re-abstracting data, we identified several reasons why conclusions changed: incorrect abstraction, improper choice of unit of analysis, use of outdated and simplistic statistical models and inadequate use of meta-regression. We describe the effect of these revised conclusions on health policy and suggest strategies to improve performance of diagnostic tests.