I had to check on this one. It is not possible to get from sensitivity/specificity to PPV/NPV without an estimate of the population prevalence of the disease. It has apparently been suggested that the PPV/NPV values be normalized assuming a 50% prevalence rate. I don't know whether 50% is an acceptable estimate of prevalence for your class. Once you have the prevalence, you should build a table like the one from Wikipedia: http://en.wikipedia.org/wiki/Sensitivity_and_specificity Populate each box with a % as follows: The first box is Positive for Condition and for Disease. Since the test has a sensitivity of 90% and 50% of the population is assumed to have the disease: $\frac{ TruePositive }{ AllTests }=\frac{ TruePositive }{ AllPositive }*\frac{ AllPositive }{ AllTests }$ $\frac{ TruePositive }{ AllTests }=Sensitivity*Prevalence(50pct)$ The formula above will work with other values for prevalence. The rest of the algebra is pretty straightforward.