Few people object when one emphasizes the importance of evaluating benefits vs. risk in the assessment of a medical intervention, but actually performing such an evaluation in a meaningful way – aye, there’s the rub! Are we talking about the effect on a population (if so, what population?) or individuals (if so, who?)? Are we referring to the totality of benefits and the totality of harms, or only some subset of one or the other? Have we measured harms sufficiently reliably to allow a fair weighing of harms to benefits or are the harms collected haphazardly and the benefits collected carefully? Are we comparing the benefit-risk profile to some competing treatment not studied concurrently with the treatment of interest to us (and, if so, do we have enough information to characterize the profile in the competing treatment?)? How useful are the metrics of NNT and NNH (number needed to treat and number needed to harm)? This talk raises a series of questions about evaluating benefit-risk, emphasizes the limitations of single metrics that purport to characterize the multivariable nature of benefits and harms, but, sadly, proposes few answers.